<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Book on Michele on Software</title><link>https://sollecitom.github.io/software-product-development-blog/tags/book/</link><description>Recent content in Book on Michele on Software</description><generator>Hugo -- gohugo.io</generator><language>en-us</language><copyright>Copyright © 2023 - Michele Sollecito</copyright><lastBuildDate>Sat, 21 Nov 2026 17:27:00 +0000</lastBuildDate><atom:link href="https://sollecitom.github.io/software-product-development-blog/tags/book/index.xml" rel="self" type="application/rss+xml"/><item><title>I Wrote a Book: Software Product Development Without Stupidity</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2026/2026-11-23-i-wrote-a-book/</link><pubDate>Sat, 21 Nov 2026 17:27:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2026/2026-11-23-i-wrote-a-book/</guid><description>&lt;p>My book is out. It&amp;rsquo;s called &lt;em>Software Product Development Without Stupidity: How Not to Sabotage Your Odds in an Already Hard Game&lt;/em>. It&amp;rsquo;s on Amazon in paperback, hardcover, and Kindle (&lt;a href="https://www.amazon.co.uk/dp/B0HSJWYBP2">UK&lt;/a>, &lt;a href="https://www.amazon.com/dp/B0HSJWYBP2">US&lt;/a>), and on &lt;a href="https://leanpub.com/without-stupidity">Leanpub&lt;/a> as a DRM-free PDF and EPUB.&lt;/p>&lt;p>If you&amp;rsquo;ve been reading this blog, you&amp;rsquo;ve seen pieces of it already. I tried to write a book several times, and every time it ended up as a handful of blog posts. Some of them grew into chapters:&lt;/p>&lt;ul>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-21-product-manager-role-is-a-mistake/">The product manager role is a mistake&lt;/a>&lt;/li>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-11-06-at-the-heart-of-the-burnout-epidemic/">At the heart of the burnout epidemic&lt;/a>&lt;/li>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-29-lifers-and-politicians/">Lifers and politicians&lt;/a>&lt;/li>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-31-so-you-think-you-work-in-a-team/">So you think you work in a team&lt;/a>&lt;/li>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-10-start-with-a-skeleton-not-to-die-a-zombie/">Start with a skeleton not to die a zombie&lt;/a>&lt;/li>&lt;li>&lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-01-introducing-abstraction-deviation-impedance/">Introducing Abstraction Deviation Impedance&lt;/a>&lt;/li>&lt;/ul>&lt;p>In the book they&amp;rsquo;re rewritten and tied into one argument.&lt;/p>&lt;h2 id="why-i-wrote-it">Why I wrote it&lt;/h2>&lt;p>Over the years I&amp;rsquo;ve talked about software product development with many people, in many places, and most of them were frustrated. They knew how to fix their company&amp;rsquo;s worst problems and weren&amp;rsquo;t allowed to. These were competent, senior people who cared about their customers and their craft. All they wanted was to do their best work, and somehow they couldn&amp;rsquo;t.&lt;/p>&lt;p>Whether it was over a pint or in a meeting room, it was always the same story. Those conversations are where this book comes from.&lt;/p>&lt;p>What finally unblocked me was Luca Debiasi, a former colleague and a friend, who at some point stopped letting me off the hook. Put it together, he said. By then, the pain of writing a book had become more tolerable than the pain of leaving heartfelt things unsaid.&lt;/p>&lt;h2 id="what-its-about">What it&amp;rsquo;s about&lt;/h2>&lt;p>Most software companies are full of smart, dedicated people. So why do the products keep getting worse, the costs keep climbing, and the people work harder than ever?&lt;/p>&lt;p>My answer is the system they work in. Competent, well-intentioned people make choices that end up harming the company, the product, and themselves, because they misunderstand the system they&amp;rsquo;re standing in. That&amp;rsquo;s what I mean by stupidity.&lt;/p>&lt;p>The book follows it from the top down: what a company is for, how people are hired, paid, and organized, how a company keeps its capability to change, and then the architecture, the code, and the practices that decide whether a product can keep changing. There&amp;rsquo;s also an interlude on AI, and on why speeding up a broken system only gets you its bad results sooner.&lt;/p>&lt;p>It&amp;rsquo;s written for founders and leaders, and it stays technical enough to be worth a developer&amp;rsquo;s time.&lt;/p>&lt;h2 id="what-it-isnt">What it isn&amp;rsquo;t&lt;/h2>&lt;p>It isn&amp;rsquo;t a manual. No seven steps to success, no frameworks, and no case studies either. I take strong positions, and some of them will irritate you. Good. A polite book full of niceties and platitudes would waste your time and mine.&lt;/p>&lt;p>I&amp;rsquo;m also not trying to convince you of anything. People don&amp;rsquo;t change their minds because someone argues harder. They change when they meet a way of working that obviously works and get curious, or when they run into a contradiction their current beliefs can&amp;rsquo;t explain. So the book shows you how I&amp;rsquo;ve come to see things, and leaves the rest to you.&lt;/p>&lt;p>If you read it, let me know what you think.&lt;/p></description></item><item><title>From Products to Obstacles Wrapped in Subscriptions: the Decline of Craftsmanship and of an Industry</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2026-01-22-obstacles-wrapped-in-subscriptions/</link><pubDate>Thu, 22 Jan 2026 08:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2026-01-22-obstacles-wrapped-in-subscriptions/</guid><description>&lt;p>We used to build products. Now all we build is obstacles wrapped in subscriptions.&lt;/p>&lt;p>Companies sell their products with an implicit promise. That they did their best to make the customer&amp;rsquo;s life better. That things will work well for the foreseeable future. That no additional unexpected costs will pop up at some point.&lt;/p>&lt;p>These days, this is a lie.&lt;/p>&lt;p>Companies don&amp;rsquo;t necessarily set out to destroy value, they just lose the ability to create it without also losing the appetite for those rewards that creating value for customers brings. They hire skilled and passionate people and then trap them in a system that prioritizes short-term capture over long-term utility and delighting customers.&lt;/p>&lt;p>It&amp;rsquo;s a betrayal of the craft, that&amp;rsquo;s what this is. And it&amp;rsquo;s a betrayal of the customer.&lt;/p>&lt;p>Look at your product. It&amp;rsquo;s not a solution. It&amp;rsquo;s a collection of exposed internal complexities that nobody had the courage to burn down. A collage of features thrown at the wall by different people, partially built, poorly integrated, and without a shred of overarching design.&lt;/p>&lt;p>We&amp;rsquo;ve become so lazy that we ask our users to do our work for us. And yet:&lt;/p>&lt;ul>&lt;li>No customer wants to edit the fields of a row in a grid to do anything.&lt;/li>&lt;li>No user wants to figure out how to navigate maze-like nested menus to unsubscribe from your product trial before it auto-renews for a year.&lt;/li>&lt;li>And definitely nobody wants to pray for ten seconds of connectivity not to lose the work they tried to do in the tube.&lt;/li>&lt;/ul>&lt;p>Somebody stated that people don&amp;rsquo;t buy a 1/4-inch drill, but that they want a 1/4-inch hole. And yet in modern Saas, we don&amp;rsquo;t give them the hole. Nor the drill. What we give them is a subscription to a configurable drill-management platform. Because our organization is too fragmented and focused on itself to ship anything else.&lt;/p>&lt;p>We force them to manually calibrate the bit and save the drill&amp;rsquo;s state every five minutes. Because strong opinions and top-down product design became dangerous in organizations often run and designed by corporate sheep.&lt;/p>&lt;p>This industry loves to claim that doing things the right way is too expensive. That we have to cut corners to stay lean.&lt;/p>&lt;p>That&amp;rsquo;s a lie.&lt;/p>&lt;p>Building it right isn&amp;rsquo;t more expensive than doing it poorly. It doesn&amp;rsquo;t require a bigger budget. It requires deep thinking and taking risks.&lt;/p>&lt;p>But modern companies have disincentivized both. Taking risks means sometimes being wrong, and being wrong is a career risk in an environment that prizes safe incremental gains over outsized discontinuous improvement. Thinking deep is quiet. It looks like inactivity to the incompetent middle manager with a Gantt chart. It doesn&amp;rsquo;t make things move up and to the right every single day.&lt;/p>&lt;p>In a system that rewards predictability, consensus, measurable progress, and visible activity, mediocrity reigns, while greatness is treated as a disease.&lt;/p>&lt;p>So we take the path of least resistance. We ship the minimum viable overhead. Because a status bar is easier to explain to a stakeholder than an automated workflow that understands human intent.&lt;/p>&lt;p>We optimized for shipping trash, because the metrics for trash are easier to track than the metrics for quality.&lt;/p>&lt;p>If your company rewards moving tickets and punishes stopping to think and failed experiments, you aren&amp;rsquo;t building a product. You&amp;rsquo;re just managing a decline.&lt;/p>&lt;p>Start shipping the reality you promised, instead of the ugliness of your corporate processes and politics. Tomorrow, choose a difficult &amp;ldquo;no&amp;rdquo; over an easy &amp;ldquo;yes&amp;rdquo;. Go back to obsessing over quality and making your customer&amp;rsquo;s life better. And, for the love of God, rid yourself of these meaningless executors that somehow wormed their way up your corporate ladder without an ounce of passion or skills.&lt;/p></description></item><item><title>Incentives, Friction, and Soul in the Game</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-10-15-incentives-friction-and-soul-in-the-game/</link><pubDate>Wed, 15 Oct 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-10-15-incentives-friction-and-soul-in-the-game/</guid><description>&lt;p>One of the things I keep thinking about is the extent to which most companies go out of their way to discourage their own employees from tackling the problems they spot.&lt;/p>&lt;p>It&amp;rsquo;s kind of crazy, right? Because that&amp;rsquo;s exactly why companies hire people in the first place. At least in theory.&lt;/p>&lt;p>Yet, the incentives and internal dynamics often push folks away from wanting to make big improvements.&lt;/p>&lt;p>You have to remember, people naturally avoid pain. And while some thrive on a bit of risk, most will steer clear of pursuits where the potential rewards are dwarfed by the downsides.&lt;/p>&lt;p>Imagine I&amp;rsquo;m a plumber, called in to fix a simple leak in your house. While I&amp;rsquo;m at it, I uncover a major plumbing disaster waiting to happen. I&amp;rsquo;d see it as my duty to flag this, and I&amp;rsquo;d speak up.&lt;/p>&lt;p>As the homeowner, you&amp;rsquo;d probably groan, more money to shell out, more hassle, maybe even no water for a couple of days?&lt;/p>&lt;p>So you&amp;rsquo;d grill me: Are you absolutely sure? What if we ignore it? You might even seek a second opinion.&lt;/p>&lt;p>Now, plumbers deal with homeowners on a one-off, transactional basis, so I have a real material interest in pointing this out. And your skepticism makes sense, because I could be inflating the issue to score a bigger job.&lt;/p>&lt;p>But software developers, even contractors, are typically paid for their time, regardless of what they accomplish day to day.&lt;/p>&lt;p>As a company employee, if I spot a problem and keep quiet, it costs me nothing. And solving a massive company issue? There&amp;rsquo;s not much personal upside, since successes get credited to the execs or the team as a whole, to sidestep any jealousy or politics.&lt;/p>&lt;p>What I&amp;rsquo;m left with is mostly just the risks and the hassle.&lt;/p>&lt;p>I&amp;rsquo;d need to flag it to the decision-makers. They&amp;rsquo;d eye me like some overeager kid itching to tinker for fun. They&amp;rsquo;d demand data and proof for stuff that&amp;rsquo;s inherently uncertain. I&amp;rsquo;d have to estimate, persuade, set milestones, track progress, and report in constantly.&lt;/p>&lt;p>I&amp;rsquo;d also have to pitch this to my colleagues, and some would feel attacked. I&amp;rsquo;d win over a few, but I&amp;rsquo;d make enemies along the way.&lt;/p>&lt;p>If it flops, even if they didn&amp;rsquo;t let me handle it my way, guess who&amp;rsquo;d take the heat? Me. I might not get canned, but people would whisper about the newbie who thought he knew it all. I&amp;rsquo;d get lectures on the sacred &amp;ldquo;domain knowledge&amp;rdquo; and how things that work elsewhere just don&amp;rsquo;t work here.&lt;/p>&lt;p>All told, I&amp;rsquo;d shoulder serious personal risk, pour in extra effort, do the heavy lifting myself, and endure the stress, for something that won&amp;rsquo;t reward me much, even if it succeeds. Doesn&amp;rsquo;t sound appealing, does it?&lt;/p>&lt;p>And that&amp;rsquo;s why most folks don&amp;rsquo;t bother.&lt;/p>&lt;p>Hold on, I can already hear you pushing back: What about skin in the game? Your bonus, equity, or profit-sharing?&lt;/p>&lt;p>Fair point, but outcomes are rarely pinned to one person. Even if I pull off something huge, my impact gets watered down or offset by other efforts. Plus, the revenues or share value of a large company isn&amp;rsquo;t the sum of its employees&amp;rsquo; contributions, so bonuses are more driven by politics than they depend on personal success. No, skin in the game doesn&amp;rsquo;t fix this.&lt;/p>&lt;p>Still, some rare individuals make it their calling to push boundaries, craft visions, overhaul broken processes, and tackle their company&amp;rsquo;s most challenging issues.&lt;/p>&lt;p>Why? Mostly because they can&amp;rsquo;t help it. That&amp;rsquo;s what true strength looks like, they pursue it even when it&amp;rsquo;s risky or inconvenient, finding self-actualization in it, seeing it as a duty to the company and themselves. In short, they have soul in the game.&lt;/p>&lt;p>Let me wrap this up by reminding you: software product development is all about power laws, exponential leaps, discovery, experimentation, and collaborative knowledge work. It&amp;rsquo;s not some predictable, assembly-line grind.&lt;/p>&lt;p>Those breakthrough moments can define a product, or an entire company.&lt;/p>&lt;p>As a leader, you want to hire mostly those unstoppable forces who chase perfection relentlessly, to guarantee real excellence.&lt;/p>&lt;p>Hiring isn&amp;rsquo;t just about whether someone can do the job, it&amp;rsquo;s about whether they will, even when the company&amp;rsquo;s setup works against them. Skills? Those can be picked up. But attitude? That seldom shifts.&lt;/p>&lt;p>Every stalled company is packed with smart folks sporting degrees and resumes who never move the needle.&lt;/p>&lt;p>Beyond hiring, acknowledge that shaking things up always involves some friction and risk. But as a company, do what you can to dial it down.&lt;/p>&lt;p>Even those with soul in the game have limits, energy drains, families to support.&lt;/p>&lt;p>Make the path too painful or perilous, and even the boldest change-makers might pull back.&lt;/p>&lt;p>So if you&amp;rsquo;re an exec, director, manager, or anyone calling shots, pay attention. Don&amp;rsquo;t foster an environment ripe for groupthink, inertia, missed opportunities, and silent disengagement.&lt;/p></description></item><item><title>Nobody Feels Responsible Without Autonomy</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-19-autonony-responsibility/</link><pubDate>Fri, 19 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-19-autonony-responsibility/</guid><description>&lt;p>Hey there, reader. If you&amp;rsquo;ve ever felt like you&amp;rsquo;re carrying the weight of the world on your shoulders but your hands are tied behind your back, this one&amp;rsquo;s for you. I&amp;rsquo;ve been in the trenches of tech and leadership long enough to see patterns that break people and companies over and over.&lt;/p>&lt;p>Today, I want to talk about something that&amp;rsquo;s burned me personally: how autonomy (or authority, if you prefer that word) and responsibility are like two sides of the same coin. Flip one without the other, and you end up with a mess.&lt;/p>&lt;p>If you&amp;rsquo;re not free to make the calls you believe in, you can&amp;rsquo;t possibly own the outcomes. It&amp;rsquo;s that simple, and yet, it&amp;rsquo;s ignored in so many organizations. Let me unpack this, share a bit of my own experience, and hopefully spark some debate.&lt;/p>&lt;h2 id="the-core-truth-no-freedom-no-ownership">The Core Truth: No Freedom, No Ownership&lt;/h2>&lt;p>Think about it like this: imagine you&amp;rsquo;re captaining a ship, but every time you try to steer away from the iceberg, the admiral on shore radios in and says, &amp;ldquo;Nah, stay the course.&amp;rdquo;&lt;/p>&lt;p>When you inevitably hit the ice, who&amp;rsquo;s getting blamed? You, the captain. That&amp;rsquo;s the reality for too many &amp;ldquo;leaders&amp;rdquo; out there. Responsibility without autonomy isn&amp;rsquo;t leadership. It&amp;rsquo;s a scapegoat setup.&lt;/p>&lt;p>In my experience, when people are given accountability but not the reins to drive change, they disengage. Why pour your soul into something when your best ideas get vetoed by someone who isn&amp;rsquo;t even in the room? I&amp;rsquo;ve felt this sting myself early in my career. After six months of pushing boulders uphill, I stopped feeling responsible. The failures weren&amp;rsquo;t mine. They were the system&amp;rsquo;s.&lt;/p>&lt;p>And honestly, that detachment hurt, as it made me question if I was even cut out for leadership. I felt anger more than once during that time, feeling like a fraud for not delivering what I knew was possible.&lt;/p>&lt;p>This isn&amp;rsquo;t just anecdotal fluff. Studies on employee engagement show that when people have autonomy, their sense of ownership skyrockets, leading to better performance and innovation. Strip it away, and you get apathy, burnout, or quiet quitting. It&amp;rsquo;s human nature. We thrive when we can shape our destiny.&lt;/p>&lt;h2 id="the-misnomer-of-modern-management">The Misnomer of Modern &amp;ldquo;Management&amp;rdquo;&lt;/h2>&lt;p>Far too often, companies hire folks into &amp;ldquo;leadership&amp;rdquo; or &amp;ldquo;management&amp;rdquo; roles, but what they really want are process administrators or glorified babysitters.&lt;/p>&lt;p>You know the type: the manager who&amp;rsquo;s stuck enforcing outdated policies, coaching underperformers without the tools to actually fix systemic issues, or just keeping the peace like a camp counselor.&lt;/p>&lt;p>There&amp;rsquo;s a place for coaching and admin work. Hell, it&amp;rsquo;s vital in support roles. But it has zero to do with true leadership.&lt;/p>&lt;p>Real management and leadership? That&amp;rsquo;s about reshaping &lt;em>how the work works&lt;/em>. It&amp;rsquo;s tinkering with processes, reallocating resources, experimenting with new approaches, and yes, sometimes rocking the boat to sail faster.&lt;/p>&lt;p>This stuff can transform a team or department. I&amp;rsquo;ve seen it firsthand: In one gig, I was given carte blanche to overhaul our CI/CD pipeline. We cut deployment times by 70%, and the team&amp;rsquo;s morale? Through the roof because they felt the wins were &lt;em>ours&lt;/em>. But without that authority, it&amp;rsquo;d have been another failed initiative gathering dust.&lt;/p>&lt;h2 id="the-classic-trap-hiring-a-hero-then-clipping-their-wings">The Classic Trap: Hiring a Hero, Then Clipping Their Wings&lt;/h2>&lt;p>Picture this scenario. It&amp;rsquo;s played out in countless companies, including ones I&amp;rsquo;ve consulted for. a business is tanking in some key area: ROI is trash, performance is sluggish, innovation is DOA. They decide, &amp;ldquo;let&amp;rsquo;s bring in a rockstar leader to fix it!&amp;rdquo; The candidate aces the interview, shares visionary ideas, gets the offer. They join, ramp up, buzzing with energy and good intentions.&lt;/p>&lt;p>Fast-forward two years: crickets. Nothing&amp;rsquo;s changed.&lt;/p>&lt;p>The new manager had killer ideas. Restructuring teams, adopting new tech, shifting priorities. But they got shot down. The folks above? Not confident in the vision, disagreed on fundamentals, couldn&amp;rsquo;t see the short-term ROI, or just didn&amp;rsquo;t want to &amp;ldquo;rock the boat.&amp;rdquo;&lt;/p>&lt;p>Classic lines: &amp;ldquo;Let&amp;rsquo;s wait and see,&amp;rdquo; or &amp;ldquo;We need more data.&amp;rdquo; Meanwhile, sticking to the broken playbook keeps yielding the same lousy results.&lt;/p>&lt;p>Why does this happen? Sometimes, companies have internalized their mediocrity as the norm. Suggesting real improvement feels like a threat to the egos of those who&amp;rsquo;ve long since given up.&lt;/p>&lt;p>Other times, the blame game kicks in. &amp;ldquo;We hired you to fix this, and it&amp;rsquo;s still broken. Your fault!&amp;rdquo; Never mind that the manager&amp;rsquo;s hands were tied. This scapegoating is soul-crushing. It erodes trust and turns talented people cynical.&lt;/p>&lt;p>And the saddest part? Often, it&amp;rsquo;s the manager themselves who bails.&lt;/p>&lt;p>Short stints on a resume aren&amp;rsquo;t ideal. They raise eyebrows in future interviews. But faking productivity while being forced into irrelevant busywork? That&amp;rsquo;s a recipe for misery. Pretending to lead when you&amp;rsquo;re just a puppet sucks the life out of you.&lt;/p>&lt;p>I&amp;rsquo;ve lost count of friends who&amp;rsquo;ve quit for this exact reason, some quietly slipping away, others venting publicly on LinkedIn or Twitter (this is not a case of venting publicly on LinkedIn :-).&lt;/p>&lt;h2 id="a-real-story-the-tech-leads-frustration">A Real Story: The Tech Lead&amp;rsquo;s Frustration&lt;/h2>&lt;p>Let me get even more personal here. Just recently, a tech lead (not someone I work with directly) slid into my LinkedIn DMs. She&amp;rsquo;s sharp, experienced, and passionate about her craft. But she confessed she&amp;rsquo;s not allowed to tackle the performance issues in her team. We&amp;rsquo;re talking quality issues that are costing the company real money and morale.&lt;/p>&lt;p>When she&amp;rsquo;s tried to step up without formal authority. Suggesting process tweaks, mentoring juniors on better practices. The results have been mixed at best, downright poor at worst. Pushback from above, resistance from peers, and no backup. She feels stuck, responsible for the team&amp;rsquo;s output but powerless to influence it meaningfully.&lt;/p>&lt;p>Hearing her story hit home because I&amp;rsquo;ve been there. It&amp;rsquo;s hard to admit, but I&amp;rsquo;ve second-guessed my own decisions in similar spots, wondering if I was the problem. Was I too aggressive? Not persuasive enough? But no. It&amp;rsquo;s the system. And stories like hers remind me why we need to call this out.&lt;/p>&lt;h2 id="breaking-the-cycle-what-can-we-do">Breaking the Cycle: What Can We Do?&lt;/h2>&lt;p>So, how do we fix this? If you&amp;rsquo;re in a position to hire or lead, start by granting real authority alongside responsibility. Define clear boundaries, but within them, let people fly. Trust their expertise. That&amp;rsquo;s why you hired them! For those feeling trapped: advocate for yourself early. In interviews, ask pointed questions like, &amp;ldquo;what&amp;rsquo;s the biggest change a recent hire made here?&amp;rdquo; If the answer is vague or outright nonexistent, run.&lt;/p>&lt;p>And to the companies out there: embrace discomfort. True growth comes from empowering people to shake things up. It might bruise some egos, but the alternative is stagnation.&lt;/p>&lt;p>I&amp;rsquo;ve learned this the hard way, through some failures and a few wins. If you&amp;rsquo;re nodding along, share your story in the comments. Let&amp;rsquo;s build a community that demands better. Because at the end of the day, autonomy isn&amp;rsquo;t a perk. It&amp;rsquo;s the fuel for real responsibility and remarkable results. What&amp;rsquo;s your take?&lt;/p></description></item><item><title>A Birthday Zoo Trip That Reminded Me Why User Experience Still Feels Like an Afterthought</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-17-user-experience-zoo/</link><pubDate>Wed, 17 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-17-user-experience-zoo/</guid><description>&lt;p>Today marks my son&amp;rsquo;s first birthday, a milestone that sneaks up on you with its quiet weight. One minute he&amp;rsquo;s a newborn bundle, the next he&amp;rsquo;s pulling himself up on the coffee table, grinning like he&amp;rsquo;s cracked a centuries old maths problem. To celebrate, my wife and I decided on the London Zoo, an easy choice, or so we thought. Animals, open spaces, a chance for him to point at a giraffe and babble something cute. But as I sat down to book the tickets online this morning, what should have been a simple step turned into a slog that nearly derailed the whole plan. It&amp;rsquo;s these moments that pull me back into thinking about how systems work, or don&amp;rsquo;t, and why so many organizations seem to forget the human at the center.&lt;/p>&lt;p>Let me walk you through it, because it&amp;rsquo;s the kind of everyday friction that builds up and shapes our choices in ways companies rarely notice. The website greeted me with a mandatory account creation screen right out of the gate. Not optional, not &amp;ldquo;guest checkout&amp;rdquo;, no, full registration required. Home address, email, phone number. Then the password dance: at least 12 characters, uppercase, lowercase, numbers, symbols. I get security matters, but for a toddler&amp;rsquo;s outing? It felt like unnecessary gatekeeping. Next, picking an entry time, 10 a.m. or noon, without much guidance on crowds or nap schedules for little ones. Payment? No Android Pay support, so I manually entered card details while juggling a half-asleep baby on my lap.&lt;/p>&lt;p>And it didn&amp;rsquo;t stop there. Checkboxes for marketing notifications (opt-in or out, and email vs SMS vs post), a nudge to donate extra for conservation, an invitation to join the Zoological Society for ongoing perks. By the time I hit submit, I&amp;rsquo;d second-guessed every input. Then came the address verification: confirm if the one I entered matched the one they auto-looked up from my details (it did, of course). Finally, three separate emails landed in my inbox, welcome aboard, order confirmation, and the actual e-tickets. All for a £60 family adventure.&lt;/p>&lt;p>On any ordinary day, I&amp;rsquo;d have closed the tab and pivoted to something easier, like a local park or even delivery pizza at home. But this was his first birthday. Special enough to push through the irritation. As I sat there afterward, tickets in my inbox, I couldn&amp;rsquo;t shake the feeling that this wasn&amp;rsquo;t just a clunky website. It was a symptom of a deeper misalignment in how the zoo, and so many places like it, thinks about their work.&lt;/p>&lt;p>From a systems thinking perspective, it&amp;rsquo;s fascinating how these processes emerge from interconnected parts that drift out of sync. The ticket system isn&amp;rsquo;t isolated but tied to marketing databases, CRM tools, integrations, and revenue streams. Someone upstream set a goal to boost registered users, maybe to track engagement or personalize future offers. That metric becomes a KPI, which cascades down to quarterly reviews and team bonuses. Meanwhile, the UX team, if there is one, inherits a Frankenstein of requirements: capture more data here, add this upsell there, ensure GDPR boxes are ticked. The result? A flow optimized for internal checkboxes, not the external reality of a parent racing against nap time. It&amp;rsquo;s a classic feedback loop gone awry, the system reinforces itself, prioritizing short-term signals over long-term relationships, until users like me quietly vote with our feet (or our wallets).&lt;/p>&lt;p>This ties straight into user experience design principles, which I&amp;rsquo;ve always approached as a practice in empathy mapping. Good UX isn&amp;rsquo;t about pretty interfaces; it&amp;rsquo;s about anticipating the mental model of the person on the other side. In this case, the zoo&amp;rsquo;s flow assumed I was a transactional buyer, methodically weighing options like a spreadsheet. But that&amp;rsquo;s not me, or most families. We&amp;rsquo;re in motion, emotions high, attention fragmented. The design should mirror that: reduce cognitive load, surface relevant choices, and build trust through clarity. Instead, it layered on decisions that felt extractive, give us your data, your preferences, a bit more money if you care. No wonder conversion rates suffer; it&amp;rsquo;s like inviting someone to dinner and making them chop the onions first.&lt;/p>&lt;p>Then there&amp;rsquo;s the jobs-to-be-done framework, which Clayton Christensen popularized and which resonates so much in moments like this. People don&amp;rsquo;t &amp;ldquo;hire&amp;rdquo; a product to fulfill abstract needs; they hire it for a specific job in a given context. What job was I hiring the London Zoo for? Not &amp;ldquo;view exotic animals,&amp;rdquo; though that&amp;rsquo;s the surface appeal. No, the real job was &amp;ldquo;create a memorable family bonding experience on a milestone day.&amp;rdquo; That could mean the thrill of a penguin waddle, sure, but also hassle-free logistics, age-appropriate pacing, and little surprises that make the day feel tailored. An aquarium might scratch a similar itch, or a backyard picnic with animal crackers. The zoo competes in that broader ecosystem, yet their process treated me like I was there for the lions alone, hence the heavy emphasis on conservation donations and memberships, as if animal advocacy was my core motivator.&lt;/p>&lt;p>If they&amp;rsquo;d started from the job, the entire system could realign. Imagine logging in with Google for instant familiarity, no password gymnastics. A quick selector for visit type: &amp;ldquo;Family Day with Toddler&amp;rdquo; versus &amp;ldquo;Solo Wildlife Enthusiast.&amp;rdquo; That triggers contextual help: estimated tour lengths (90 minutes for the kid-friendly loop?), animal activity peaks (monkeys perk up post-noon feedings?), and practical reassurances (yes to packed lunches; start at the east gate near the pram-friendly Tube exit). Doubts dissolve with a glance. Then, thoughtful upsells emerge naturally: a £5 birthday badge for the little one, bundled snacks to skip the queues, or a photo package capturing that first giraffe gasp. Checkout? One-tap with Android Pay. Single email: tickets, map highlights, and a &amp;ldquo;Pro Tip: Pack extra wipes&amp;rdquo; nudge. Friction minimized, value amplified.&lt;/p>&lt;p>Building this wouldn&amp;rsquo;t just be kinder; it&amp;rsquo;d be smarter. Systems thinking tells us that simpler inputs yield more resilient outputs, easier to iterate, test, and scale. UX design ensures it delights at every touchpoint, turning one-time buyers into advocates. And jobs theory? It uncovers untapped revenue: families willing to pay premiums for personalization, not because they&amp;rsquo;re generous, but because it nails the emotional payoff.&lt;/p>&lt;p>The London Zoo makes for a stark case study in how organizational silos erode user-centricity, marketing chasing lists, ops enforcing rules, execs eyeing metrics, but it&amp;rsquo;s hardly alone. We&amp;rsquo;ve all bounced off similar walls, from booking flights to signing up for newsletters. So, as I wrap up this reflection ahead of our zoo day (fingers crossed for no meltdowns), I keep circling back to one question: What&amp;rsquo;s the real cure for this epidemic of misaligned systems? Is it leadership that champions cross-functional empathy? Tools that simulate user journeys end-to-end? Or something more cultural, like tying incentives directly to lifetime value over quarterly wins? I&amp;rsquo;d love to hear your take, what&amp;rsquo;s worked in your world to keep the human job front and center?&lt;/p></description></item><item><title>The Mirror Never Lies: An Ode to Unbroken Rebels</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-15-the-mirror-never-lies/</link><pubDate>Mon, 15 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-15-the-mirror-never-lies/</guid><description>&lt;p>Every morning, I drag myself to the sink, razor in hand, steam curling like a ghost around the fogged glass. The water burns, the blade stings, and for those raw seconds, it’s just me and the truth staring back.&lt;/p>&lt;p>No likes, no clout, no roles, and no polished mask to hide behind. Just my soul and one question: Who are you when no one’s watching?&lt;/p>&lt;p>The world has its playbook, and it’s a bestseller. Write what sells. Chase crypto scams, echo political outrage, praise metrics, or preach AI salvation, whatever trends in the endless scroll.&lt;/p>&lt;p>Tempting, right? Bury your soul, amplify the noise, and watch opportunities stack up like chips on a poker table. Why bleed for your beliefs when you can mimic the mob, toss in buzzwords, and cash in? Smooth your edges, nod to the crowd, and you’re set: doors open, DMs buzz with deals, the siren call of “making it.”&lt;/p>&lt;p>It’s a simple trade: your spine for a seat at a hollow table. I’ve felt that pull, God I still do.&lt;/p>&lt;p>Standing at the cliff’s edge, staring at a blank page, fingers itching to type the words that’d win the room, the gig, the fleeting rush of being “somebody.”&lt;/p>&lt;p>Opportunities dangle like overripe fruit: a viral retweet, a sketchy brand deal, some disinvested advice, a chance to ride the next wave of nonsense. All it takes is a small betrayal.&lt;/p>&lt;p>But every time I lean in, something in me fights back, a fire in my gut, scars on my knuckles, refusing to let me forget who I am.&lt;/p>&lt;p>My beliefs aren’t just opinions to swap like cheap sneakers. They’re the iron in my spine, the grit in my veins, the drumbeat driving me since I decided my voice was worth the fight. Strip them away, and I’m nothing, just another empty hustler grinning for the algorithm, peddling recycled platitudes for pennies.&lt;/p>&lt;p>I’d miss the deals, the keynotes and conferences, the slick promise of exposure. But losing the person in the mirror? That’s a ruin no spotlight can fix.&lt;/p>&lt;p>The world’s a circus of grifters, seats sold out to chameleons with eyes like cash registers. They flip from wellness gurus to crypto gurus to AI prophets, whatever pays fastest.&lt;/p>&lt;p>Some are sharp, spinning opportunism into gold; others are shadows, flailing for scraps because they never stood for anything. Let them have their hustle. Let them wake at forty, faces creased with regret, staring at the flicker they traded for fire.&lt;/p>&lt;p>Not us. Not the holdouts, the heretics too stubborn to bend. We’re no heroes, just battered souls writing what burns us raw, speaking truths that sting the tongue, living with a courage that makes the mirror proud.&lt;/p>&lt;p>We miss out, and it hurts: every ignored post, every slammed door, every empty inbox stings.&lt;/p>&lt;p>But we show up, whole, unbowed, jaded but blazing, because the alternative is a life choking on fakes. In a world that rewards cowards, that’s the only currency that matters: our wild, beating hearts, unapologetic and alive.&lt;/p>&lt;p>So here’s to you, the unbroken rebels, the weary dreamers who stare down the mirror and choose the harder path. This is our anthem, our roar: We won’t sell our souls for silver. We won’t dim our light for likes. We won’t trade our truths for coins or applause.&lt;/p>&lt;p>You, with your ink-stained hands and bruised convictions, you’re the pulse of a world that’s forgotten how to feel, the spark that sets it ablaze. Keep standing. Keep swinging. Keep burning. The opportunities that matter will find you, drawn like moths to the flame you refuse to snuff. And when they do, you’ll walk through those doors not as a shadow, but as a force, carrying every truth you never betrayed.&lt;/p>&lt;p>To the defiant, the pissed-off, the ones who’d rather break than kneel: You’re the rebellion the world needs. Hold fast. The mirror’s a bitch, but it’s the only one telling the truth.&lt;/p>&lt;p>(A nod to Barry O’Reilly&amp;rsquo;s and Marco Consolaro&amp;rsquo;s recent sparks: your words fanned these flames. Keep roaring!)&lt;/p></description></item><item><title>Fork, Don't Merge: Why I Prefer A Variety of Options Over Early Agreement</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-11-fork-dont-merge/</link><pubDate>Thu, 11 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-11-fork-dont-merge/</guid><description>&lt;p>Today, I want to share my concerns about a trend I’ve noticed in many companies: jumping to agreement too quickly when debating solutions.&lt;/p>&lt;p>It feels like teamwork, but I’ve seen it dull good ideas and lead to weak results. Instead, I’ve found that forking ideas early, letting them grow separately before blending them, works better.&lt;/p>&lt;p>Here’s why I prefer this approach, based on what I’ve learned from working with teams.&lt;/p>&lt;h2 id="why-early-agreement-sucks">Why Early Agreement Sucks&lt;/h2>&lt;p>I’ve been in meetings where ideas are discussed early and consensus drives what options are further explored, and which aren&amp;rsquo;t.&lt;/p>&lt;p>The intent is good, and everyone wants to feel heard, but it smooths out the bold edges that make ideas stand out. It’s like mixing everything together before the pieces have a chance to take shape. You end up with something vague and hard to build on.&lt;/p>&lt;p>In my experience, this approach muddies the direction that great work needs. A strong plan or product comes from a powerful vision, not a patchwork of compromises. When every step needs group approval, the result feels like leftovers from a negotiation, not something purposeful.&lt;/p>&lt;p>The best breakthroughs I’ve seen usually come from one person’s bold idea, not a group averaging out opinions.&lt;/p>&lt;p>Plus, focusing on agreement early can make you miss the bigger picture. Teams get stuck arguing over details and lose sight of how they fit together. I’ve had moments in projects where I realized too late that stepping back could have sparked a better solution.&lt;/p>&lt;p>And when you mix in multiple senior people, consensus-driven approaches often derail entirely. Strong personalities and competing priorities clash, stalling progress and diluting ideas further. Things take forever, progress is painful, and meetings soon start to feel meaningless.&lt;/p>&lt;h2 id="why-forking-ideas-works-better">Why Forking Ideas Works Better&lt;/h2>&lt;p>Instead, I’ve learned to encourage teams to fork ideas early on. Let people explore different angles, theories, or prototypes separately, giving each one room to grow.&lt;/p>&lt;p>This approach has worked for me time and time again. When you create a range of options, you’re more likely to find something truly strong. It turns brainstorming into a space for bold ideas, quick experiments, or deep dives into possibilities others might overlook.&lt;/p>&lt;p>Bringing in senior people here is a strength: their diverse experiences and perspectives lead to a richer variety of forks, each with unique depth.&lt;/p>&lt;p>Once those forks have developed, share them with a decision-maker. They can review the options, pick the best one (or combine a few if it fits), explain the choice, and get everyone on board. Blending only happens here, after real exploration, so the final direction builds on the best of all paths.&lt;/p>&lt;h2 id="keeping-options-open-for-better-decisions">Keeping Options Open for Better Decisions&lt;/h2>&lt;p>What I love about this is how it keeps all ideas alive until the responsible person sees them. Leaders often have a broader view, shaped by strategy or instinct, that teams might not consider. If you narrow things down too soon to reach agreement, those perspectives miss out. I’ve kicked myself before for dropping a risky idea early, only to realize later it could have been perfect for the bigger goal.&lt;/p>&lt;p>By producing and sharing a wide set of ideas, you let leaders make thoughtful choices. They can see connections, weigh trade-offs, and pick with confidence. It also builds trust, teams know their work reached the top without being watered down, which makes the final plan feel fair and exciting.&lt;/p>&lt;h2 id="how-leaders-can-encourage-variety">How Leaders Can Encourage Variety&lt;/h2>&lt;p>Sometimes, though, I’ve noticed leaders push for being presented a single agreed-upon option. I think this can come from a desire to avoid blame: if a decision goes wrong, they can point to the group’s consensus and say it was the only reasonable choice. Having multiple options challenges that safety net, but it leads to better decisions.&lt;/p>&lt;p>As someone who’s led teams, I’ve learned it’s my job to ask for variety. I don’t just want good ideas; I want a wide range of them, different in risk, style, and scope. Holding teams to that standard sparks creativity that doubt or group pressure can squash.&lt;/p>&lt;p>If I even hint that I prefer safe ideas, people will hold back and play it safe, cutting out bold options to avoid pushback. That cautious approach has killed more great ideas than I’d like to admit. Instead, I try to make forking the norm: celebrate unusual and bold ideas, give them space, and see what new possibilities emerge.&lt;/p>&lt;p>Ultimately, though, if a company makes it unsafe for a decision maker to sometimes be wrong, they can only be expected to play it safe and avoid bold proposals. Everybody is self-serving, so companies should ensure the right alignments are in place, not to encourage errors of omission only to avoid errors of commission.&lt;/p>&lt;h2 id="when-agreement-makes-sense">When Agreement Makes Sense&lt;/h2>&lt;p>I’m not against agreement, it’s valuable, just not at the start. Save it for the end, after you’ve explored the forks and chosen a path. Then, it’s great for getting everyone aligned and moving forward smoothly, built on a foundation of deep thinking.&lt;/p>&lt;p>In my experience, forking ideas early, waiting to judge them until the end, and then committing fully leads to stronger results. It respects the original spark, uses leadership’s insight, and avoids the drag of constant compromise.&lt;/p>&lt;p>Next time your team faces a big decision, try letting the ideas diverge for a while. You might be surprised by how much stronger the outcome is.&lt;/p></description></item><item><title>Mastering Software Craftsmanship: Making the Change Easy, Then Making the Easy Change</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-06-make-the-change-easy/</link><pubDate>Sat, 06 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-06-make-the-change-easy/</guid><description>&lt;p>Kent Beck’s timeless advice, &lt;em>“Make the change easy, and then make the easy change,”&lt;/em> is a guiding principle for software craftsmanship.&lt;/p>&lt;p>It’s a simple yet profound idea that should resonate with both developers writing code and leaders shaping technical strategy. At its core, this philosophy advocates for intentional preparation to ensure changes to a codebase are smooth, predictable, and sustainable.&lt;/p>&lt;p>But what does “easy” really mean, and how can we apply this principle across different dimensions of software development?&lt;/p>&lt;p>Let’s explore this concept, its practical applications, and why it matters to both developers and technical leaders.&lt;/p>&lt;h2 id="the-essence-of-make-the-change-easy">The Essence of “Make the Change Easy”&lt;/h2>&lt;p>The structure of your code can make or break the ease of implementing new behavior.&lt;/p>&lt;p>A well-organized codebase turns a potentially complex change into a straightforward task, while a tangled one can transform even a small tweak into a nightmare.&lt;/p>&lt;p>Beck’s advice encourages us to proactively reshape the code’s structure to simplify the intended change &lt;em>before&lt;/em> diving into the implementation.&lt;/p>&lt;p>This preparation reduces risk, saves time, and enhances the quality of the outcome.&lt;/p>&lt;p>However, “easy” isn’t just about the code itself. It’s about the people, processes, and tools involved.&lt;/p>&lt;p>For developers, it means writing maintainable code and reducing friction in their workflows. For directors and CTOs, it’s about fostering an environment where teams can deliver reliably and efficiently.&lt;/p>&lt;p>Let’s look at some ways to do this.&lt;/p>&lt;h2 id="practical-ways-to-make-change-easy">Practical Ways to Make Change Easy&lt;/h2>&lt;h3 id="write-tests-to-clarify-intent">Write Tests to Clarify Intent&lt;/h3>&lt;p>Writing a test that showcases the desired behavior is a powerful way to make a change easier.&lt;/p>&lt;p>A well-crafted test acts as a blueprint, guiding you toward the exact outcome you want. It also serves as a safety net, confirming when the change is complete and correct.&lt;/p>&lt;p>For developers, this means less guesswork and faster feedback.&lt;/p>&lt;p>For leaders, it translates to higher confidence in the team’s ability to deliver features without introducing bugs.&lt;/p>&lt;h3 id="build-a-fast-comprehensive-test-suite">Build a Fast, Comprehensive Test Suite&lt;/h3>&lt;p>A robust test suite that runs quickly is a game-changer.&lt;/p>&lt;p>It allows developers to verify changes in seconds, catching regressions before they reach production.&lt;/p>&lt;p>For CTOs, a fast test suite means shorter development cycles and lower risk of costly errors. Investing in automated testing infrastructure is a strategic decision that pays dividends in agility and reliability.&lt;/p>&lt;h3 id="document-clearly-and-create-executable-specifications">Document Clearly and Create Executable Specifications&lt;/h3>&lt;p>Good documentation and executable specifications (like behavior-driven development artifacts) make future changes easier for everyone.&lt;/p>&lt;p>Clear documentation reduces onboarding time for new developers and minimizes confusion during maintenance. Executable specifications align developers and stakeholders, ensuring everyone understands the system’s behavior.&lt;/p>&lt;p>For directors, this means less technical debt and smoother collaboration across teams.&lt;/p>&lt;h3 id="embrace-small-steps-and-safe-deployments">Embrace Small Steps and Safe Deployments&lt;/h3>&lt;p>Breaking changes into small, incremental steps reduces risk and simplifies debugging.&lt;/p>&lt;p>Techniques like canary rollouts and feature flags allow you to deploy changes gradually, monitoring their impact in real time.&lt;/p>&lt;p>For developers, this means less stress during deployments. For leaders, it ensures business continuity and minimizes customer-facing issues.&lt;/p>&lt;h3 id="invest-in-observability">Invest in Observability&lt;/h3>&lt;p>Proper logging and metrics make it easier to operate and troubleshoot a codebase.&lt;/p>&lt;p>When something goes wrong, well-designed observability tools help developers pinpoint issues quickly, reducing downtime.&lt;/p>&lt;p>For CTOs, this translates to better system reliability and happier customers.&lt;/p>&lt;p>Prioritize logging key events and exposing meaningful metrics to ensure your system is transparent and manageable.&lt;/p>&lt;h3 id="level-up-your-skills">Level Up Your Skills&lt;/h3>&lt;p>Sometimes, the barrier to an easy change isn’t the code but your own knowledge.&lt;/p>&lt;p>If a task feels daunting (say, writing a complex SQL query), invest time in learning. An hour spent improving your skills can make a once-difficult change trivial.&lt;/p>&lt;p>For developers, this builds confidence and expertise. For leaders, encouraging continuous learning creates a more capable, adaptable team.&lt;/p>&lt;h2 id="why-this-matters-for-developers-and-leaders">Why This Matters for Developers and Leaders&lt;/h2>&lt;p>For developers, Beck’s principle is a call to prioritize clarity, simplicity, and foresight in their work. It’s about writing code that’s not just functional but also maintainable and extensible. By making changes easy, developers reduce their own stress and create a codebase that’s a joy to work with.&lt;/p>&lt;p>For directors and CTOs, this philosophy aligns with strategic goals: delivering value quickly, minimizing technical debt, and building resilient systems. By fostering a culture that values preparation through testing, documentation, and observability, leaders empower their teams to innovate without fear of breaking things.&lt;/p>&lt;h2 id="putting-it-into-practice">Putting It Into Practice&lt;/h2>&lt;p>“Make the change easy, and then make the easy change” is more than a catchy phrase—it’s a mindset that gradually transforms how we approach software development.&lt;/p>&lt;p>Whether you’re a developer refactoring code or a CTO shaping your team’s processes, this principle encourages proactive problem-solving. Here’s how you can start:&lt;/p>&lt;ul>&lt;li>&lt;p>&lt;strong>Developers&lt;/strong>: Next time you face a complex change, pause to ask, “How can I make this easier?” Write a test, refactor a messy module, or add a feature flag.&lt;/p>&lt;/li>&lt;li>&lt;p>&lt;strong>Leaders&lt;/strong>: Invest in tools and practices that reduce friction, like automated testing or observability platforms. Encourage your team to prioritize preparation over quick fixes.&lt;/p>&lt;/li>&lt;/ul>&lt;h2 id="your-turn-how-do-you-make-change-easy">Your Turn: How Do You Make Change Easy?&lt;/h2>&lt;p>What strategies do you use to simplify changes in your codebase? Do you go out of your way to make things easy, or do tight deadlines push you toward shortcuts? Share your thoughts and experiences. Whether you’re a developer or a technical leader, there’s always a new way to apply this heuristic.&lt;/p>&lt;p>By embracing Kent Beck’s wisdom, we can all build software that’s not just functional but also a pleasure to maintain and evolve. Let’s make the change easy, together.&lt;/p></description></item><item><title>Introducing Abstraction Deviation Impedance</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-01-introducing-abstraction-deviation-impedance/</link><pubDate>Mon, 01 Sep 2025 01:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-09-01-introducing-abstraction-deviation-impedance/</guid><description>&lt;h2 id="the-search-for-a-better-term">The Search for a Better Term&lt;/h2>&lt;p>Over the years, leading teams and advising executives, I&amp;rsquo;ve often used &amp;ldquo;technical debt&amp;rdquo; to describe those persistent issues that slow us down or make changes risky.&lt;/p>&lt;p>I loved the analogy at first, and it felt like an effective way to flag problems in our codebases or designs. But after countless discussions with developers and stakeholders from all backgrounds, I realised the term creates more problems than it solves.&lt;/p>&lt;p>That’s why I’ve been working on a different way to frame these situations: Abstraction Deviation Impedance.&lt;/p>&lt;h2 id="why-technical-debt-falls-short">Why &amp;ldquo;Technical Debt&amp;rdquo; Falls Short&lt;/h2>&lt;p>Let me explain why &amp;ldquo;technical debt&amp;rdquo; doesn’t work well.&lt;/p>&lt;p>The word &amp;ldquo;technical&amp;rdquo; suggests the problems are rooted in the technology itself, be it the code, our tools, or the way the infrastructure works. In my experience, though, the real issues often lie in our models and abstractions: how we’ve conceptualized the domain, structured components, and managed complexity. These aren’t just tech issues; they’re about how well our conceptual models align with business needs and their operating environment.&lt;/p>&lt;p>The &amp;ldquo;debt&amp;rdquo; part is even more problematic. It implies something borrowed deliberately, like a loan to gain speed now with a plan to repay later. That metaphor sounds strategic, which is why it’s appealing, but it lets teams off the hook too easily. I’ve seen developers use &amp;ldquo;technical debt&amp;rdquo; to justify any mess they’ve made or dire state the system is in, framing it as a considered business strategy or calculated risk when, often, it was just neglect, rushed work, a lack of clarity, or a failure to adjust to moving needs.&lt;/p>&lt;p>In reality, these issues usually emerge unintentionally and undetected, from evolving requirements and small oversights that snowby. It&amp;rsquo;s rarely a calculated trade-off, and the business is often unaware.&lt;/p>&lt;p>Unlike a loan with predictable interest, these issues create unpredictable risks and costs.&lt;/p>&lt;h2 id="in-defence-of-technical-debt">In Defence of Technical Debt&lt;/h2>&lt;p>In its defence, the term &amp;ldquo;technical debt&amp;rdquo; was never meant to be used the way people frequently use it these days.&lt;/p>&lt;p>However, the damage it causes is real.&lt;/p>&lt;p>With its analogy with financial debt, which fuels growth and can be a solid decision if you plan to repay it, people use &amp;ldquo;technical debt&amp;rdquo; to justify a perpetual state of short-term thinking and tactical solutions.&lt;/p>&lt;p>Abstraction Deviation Impedance isn’t here to replace Technical Debt entirely. That term still works for deliberate, short-term trade-offs where you knowingly take on a cost.&lt;/p>&lt;p>The new term addresses a broader, often unintentional situation where the system’s abstractions have drifted, creating persistent friction no matter how we got there. It’s about describing the state of a system when the challenge isn’t just repaying a loan but navigating an ongoing, systemic drag that no one signed up for.&lt;/p>&lt;h2 id="defining-abstraction-deviation-impedance">Defining Abstraction Deviation Impedance&lt;/h2>&lt;p>So, what is Abstraction Deviation Impedance? Let’s break it down.&lt;/p>&lt;h3 id="abstraction-the-core-of-software-design">Abstraction: The Core of Software Design&lt;/h3>&lt;p>&amp;ldquo;Abstraction&amp;rdquo; refers to the simplifications we create to manage complexity, including domain models, APIs, and data structures.&lt;/p>&lt;p>These are the mental and architectural frameworks that make our systems understandable and adaptable.&lt;/p>&lt;p>When done right, abstractions enable smooth delivery and evolution, aligning closely with what the business needs while remaining capable of reacting to changes.&lt;/p>&lt;h3 id="deviation-the-gap-between-ideal-and-actual">Deviation: The Gap Between Ideal and Actual&lt;/h3>&lt;p>&amp;ldquo;Deviation&amp;rdquo; captures the difference between the ideal and the actual.&lt;/p>&lt;p>The ideal is a set of abstractions that perfectly reflect the problem domain: clean, consistent, and easy to extend.&lt;/p>&lt;p>In practice, deviations show up in many ways. Sometimes, it’s incorrect modeling from the start, like a payment system that assumes fixed transaction types and chokes on new ones.&lt;/p>&lt;p>Other times, it’s a gradual drift, where quick fixes or patches pile up, twisting abstractions into something convoluted.&lt;/p>&lt;p>Or it could be moving operating conditions, like a new user patterns or compliance rules, that leave once-fit models outdated.&lt;/p>&lt;p>This gap isn’t about pointing fingers; it emerges through iterations, team changes, or shifting priorities.&lt;/p>&lt;p>It’s the distance between what we envisioned and what we have, and this distance creates problems.&lt;/p>&lt;h3 id="impedance-resistance-to-flow">Impedance: Resistance to Flow&lt;/h3>&lt;p>&amp;ldquo;Impedance&amp;rdquo; is what I find most compelling.&lt;/p>&lt;p>In electrical engineering, impedance measures how a circuit resists the flow of current.&lt;/p>&lt;p>In software, it describes resistance to the flow of value, whether that’s delivering features, incorporating feedback, or scaling operations.&lt;/p>&lt;p>When abstractions deviate from the ideal, every change meets more resistance: debugging takes longer, onboarding new people gets harder, and small updates risk unexpected side effects.&lt;/p>&lt;p>This isn’t a one-time hit but a constant drag that raises costs, increases risks, and can frustrate teams as their efforts feel stuck.&lt;/p>&lt;h2 id="nothing-is-perfect">Nothing is Perfect&lt;/h2>&lt;p>Abstraction Deviation Impedance not only describes the problem well, but invites people to ask the right questions and discuss interventions.&lt;/p>&lt;p>It’s not a perfect term, but it captures the nuances I’ve grappled with over years of building and maintaining software.&lt;/p>&lt;p>It sidesteps the trap of dressing up messy systems as smart borrowing and focuses on the structural reality we face.&lt;/p>&lt;p>If you’re wrestling with similar challenges and struggle to convey the danger of the situation, try it in your next roadmap discussion or retrospective.&lt;/p>&lt;p>Let me know how it resonates, I’m curious to hear your take.&lt;/p></description></item><item><title>AI-Powered Dual-Track Product Development Before PMF</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-08-28-ai-dual-track-startup/</link><pubDate>Thu, 28 Aug 2025 19:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-08-28-ai-dual-track-startup/</guid><description>&lt;p>I wonder whether AI can enable a dual track model for building products while finding product market fit.&lt;/p>&lt;h2 id="the-historical-problem">The Historical Problem&lt;/h2>&lt;p>So far, companies were forced to choose between speed of validation and iteration, and stability, scalability, and maintainability.&lt;/p>&lt;p>And so most companies choose speed of prototyping, hoping to fix these gaps after finding PMF.&lt;/p>&lt;p>A few succeed, while the majority fail, because by then the cost of addressing these issues becomes massive, and rewriting from scratch is always unthinkable. Poor internal abstractions and quality lead to massive slowdowns, sometimes even before PMF is found.&lt;/p>&lt;h2 id="ai-to-the-rescue">AI to the Rescue?&lt;/h2>&lt;p>Now, with AI prototyping is quicker than ever, but the quality problems remain, and AI doesn&amp;rsquo;t seem to help much with these.&lt;/p>&lt;p>I wonder whether a dual track approach could now work, where fast prototyping enabled by AI is paired with building a sustainable system in parallel.&lt;/p>&lt;p>Splitting your employees among these two work streams isn&amp;rsquo;t a new idea, but with AI the effort required to prototype could be significantly lower, leaving more capacity to build solid foundations.&lt;/p>&lt;h2 id="hopes-for-a-better-future">Hopes for a Better Future&lt;/h2>&lt;p>I might be dreaming here, but it feels like it should work well. Whether the industry will ever recognize the importance of internal quality is another story.&lt;/p>&lt;p>What do you think? Does it make sense? Do you see this happening?&lt;/p></description></item><item><title>Why Your Codebase Fights Change (And How to Fix It)</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-08-25-why-your-codebase-fights-change/</link><pubDate>Mon, 25 Aug 2025 10:35:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2025/2025-08-25-why-your-codebase-fights-change/</guid><description>&lt;p>If you’ve worked with software for any length of time, you’ve seen it: a codebase that starts as a lean, focused system can morph into a costly, resistant beast without careful stewardship. I recently led a workshop on this topic, and it sharpened my perspective on what drives long-term success in development. What&amp;rsquo;s difficult isn&amp;rsquo;t implementing algorithms or optimizing data structures. It’s about ensuring that ongoing changes remain affordable, low-risk, and efficient. For those steering teams through the pressures of tight deadlines, growing systems, and collaborative demands, managing complexity is what keeps projects on track, budgets in check, and teams productive.&lt;/p>&lt;h2 id="what-development-really-entails">What Development Really Entails&lt;/h2>&lt;p>At its core, software development is about adapting a system’s behavior to meet new needs. This happens continuously, in response to the business chasing opportunities or addressing risks. Think of changes in behaviour as a series of decisions: when a certain event occurs, if specific conditions hold, then execute a particular action. An example is &amp;ldquo;when a transaction completes, if a user has enabled this, then they receive a notification that this happened&amp;rdquo;.&lt;/p>&lt;p>Simple in theory, but the same change can be a quick win in one codebase and a costly slog in another. Why? Because the work isn’t just coding the new logic, but also ensuring that the rest of the system continues to function as expected.&lt;/p>&lt;p>Every change requires:&lt;/p>&lt;ol>&lt;li>Identifying the trigger points where the update applies.&lt;/li>&lt;li>Evaluating conditions, often entangled with state management.&lt;/li>&lt;li>Implementing the new behavior.&lt;/li>&lt;li>Verifying the change delivers as intended.&lt;/li>&lt;li>Confirming existing functionality remains intact.&lt;/li>&lt;/ol>&lt;p>Complexity is what turns these steps into a grind, especially when multiple contributors are involved. Left unchecked, it inflates timelines, escalates risks, and drains resources. The real challenge of software development is keeping complexity bounded as the system evolves, ensuring teams can iterate without sinking into a maintenance nightmare.&lt;/p>&lt;h2 id="understanding-complexitys-causes">Understanding Complexity’s Causes&lt;/h2>&lt;p>Complexity isn’t an abstract concept, but the friction that slows teams down and spikes costs. It stems from three key factors, all tied to how code is structured:&lt;/p>&lt;ol>&lt;li>Dependencies: Which can make a small change ripple unpredictably across the system. You want a banana, but it comes with a gorilla, which in turn comes with the jungle.&lt;/li>&lt;li>Cognitive load: The amount of things you must juggle in your mind to operate the change.&lt;/li>&lt;li>Unknown unknowns: Hidden behaviors or edge cases that surface at the worst possible moments.&lt;/li>&lt;/ol>&lt;p>These factors don’t scale linearly. A system twice as large is typically more than twice as complex. Modularity is the antidote, breaking code into smaller, self-contained units to limit the scope of any single change. A module, whether a function, class, or library, has a contract (what it does) and an implementation (how it does it). A strong contract hides the “how,” using clear specifications like types to minimize vague expectations.&lt;/p>&lt;p>Modules deliver by:&lt;/p>&lt;ol>&lt;li>Reducing cognitive load: Teams only need to grasp the contract to use a module, not its guts unless you’re changing it.&lt;/li>&lt;li>Clarifying dependencies: Explicit connections make systems easier to reason about and maintain.&lt;/li>&lt;li>Containing changes: Localized updates prevent cascading side effects, protecting stability.&lt;/li>&lt;/ol>&lt;h2 id="shifting-to-strategic-development">Shifting to Strategic Development&lt;/h2>&lt;p>It’s easy to prioritize speed and churn out features to hit deadlines. But that short-term speed leads to long-term slowdowns. Strategic development emphasizes structures that make future changes cheaper and safer. Every tweak to behavior is a chance to refine the system’s structure and architecture. Without care, you’re stacking up complexity that inflates future costs, delays market delivery, and erodes competitive edge.&lt;/p>&lt;p>Below are some development principles, grouped for clarity, to guide a strategic approach drawn from real-world experience:&lt;/p>&lt;h3 id="organizing-for-clarity">Organizing for Clarity&lt;/h3>&lt;p>Clear, intentional structure reduces confusion and speeds up development.&lt;/p>&lt;ol>&lt;li>Structure the codebase as a graph of modules with clear dependencies, avoiding chaotic, flat namespaces.&lt;/li>&lt;li>Eliminate circular dependencies, cycles between files or modules, because they create fragility that slows progress.&lt;/li>&lt;li>Name elements after business or domain concepts, not technical specifics, for clarity.&lt;/li>&lt;li>Balance high cohesion (grouping related logic) with low coupling (reducing inter-module reliance).&lt;/li>&lt;li>Ensure APIs and structures are discoverable without deep investigation. As an example, avoid static functions, and inject dependencies through constructors.&lt;/li>&lt;/ol>&lt;h3 id="reducing-risks">Reducing Risks&lt;/h3>&lt;p>Proactive design choices minimize errors and unexpected impacts.&lt;/p>&lt;ol>&lt;li>Maximize formal requirements (like those enforced by a type system) to reduce ambiguous expectations.&lt;/li>&lt;li>Design function signatures to accommodate new data without forcing widespread updates.&lt;/li>&lt;li>Request all potential decision-making data upfront, not just what’s needed today, to avoid rework.&lt;/li>&lt;li>Build and test against module contracts, not internal details, to decouple components.&lt;/li>&lt;li>Write intuitive code that minimizes surprises for team members. Avoid surprises, and use sensible defaults that work in most cases.&lt;/li>&lt;li>Document informal requirements thoroughly to eliminate guesswork. In particular, document contracts extensively with their preconditions, checks, thrown errors, etc.&lt;/li>&lt;/ol>&lt;h3 id="enabling-efficient-changes">Enabling Efficient Changes&lt;/h3>&lt;p>Processes and patterns can make modifications predictable and safe.&lt;/p>&lt;ol>&lt;li>Simplify logic with patterns like polymorphism or factories instead of nested conditionals.&lt;/li>&lt;li>Start with a test demonstrating the desired behavior, then work backward to contract and implementation.&lt;/li>&lt;li>Define contracts and interactions before coding the details.&lt;/li>&lt;li>Break changes into small, incremental steps to minimize risk.&lt;/li>&lt;li>Adopt clear packaging patterns, like Ports and Adapters, to make intent transparent.&lt;/li>&lt;/ol>&lt;h3 id="strengthening-team-practices">Strengthening Team Practices&lt;/h3>&lt;p>Workflows and approaches can prevent entire classes of issues.&lt;/p>&lt;ol>&lt;li>Test based on dependency chains. Retest dependent modules when their dependencies change, not vice versa. If A depends on B, test A when B changes, changes in A shouldn&amp;rsquo;t require retesting B.&lt;/li>&lt;li>Restrict module contexts, like keeping database logic out of API endpoints, to prevent misuse.&lt;/li>&lt;li>Use practices like Test-Driven Development, Domain-Driven Design, Event Sourcing, or Continuous Integration to make entire categories of errors impossible.&lt;/li>&lt;li>Embrace the mindset: make the change easy, and then make the easy change (as Kent Beck eloquently put it).&lt;/li>&lt;/ol>&lt;h2 id="implications-for-leadership">Implications for Leadership&lt;/h2>&lt;p>Managing complexity isn’t just a coding problem, but a strategic imperative that shapes processes, team dynamics, and business outcomes. Uncontrolled complexity can enormously increase maintenance costs, delay market delivery, or erode customer satisfaction, undermining your competitive edge. Here’s how to apply these principles in practice:&lt;/p>&lt;h3 id="process-alignment">Process Alignment&lt;/h3>&lt;ol>&lt;li>Embed modularity and strategic principles into workflows.&lt;/li>&lt;li>Encourage small, incremental changes and rigorous testing to catch issues early.&lt;/li>&lt;li>Use Continuous Integration pipelines to enforce dependency rules and contract-based testing, reducing regressions.&lt;/li>&lt;/ol>&lt;h3 id="team-empowerment">Team Empowerment&lt;/h3>&lt;ol>&lt;li>Foster a culture where developers prioritize long-term maintainability over quick fixes.&lt;/li>&lt;li>Provide training on modular design and tools like static analysis to catch complexity creep.&lt;/li>&lt;li>Set clear documentation standards for informal requirements to save countless hours of confusion.&lt;/li>&lt;/ol>&lt;h3 id="ditch-arbitrary-deadlines">Ditch Arbitrary Deadlines&lt;/h3>&lt;ol>&lt;li>Avoid blanket deadlines, as most initiatives don’t have sharp cost-of-delay curves.&lt;/li>&lt;li>Recognize that pressure leads to shortcuts and neglect of structure—slightly quicker today, significantly slower tomorrow.&lt;/li>&lt;/ol>&lt;h3 id="resource-allocation">Resource Allocation&lt;/h3>&lt;ol>&lt;li>Invest in refactoring and technical debt reduction as regular work, not an afterthought.&lt;/li>&lt;li>Budget time for defining contracts and structuring codebases to prevent exponential cost increases later.&lt;/li>&lt;/ol>&lt;h3 id="performance-management">Performance Management&lt;/h3>&lt;ol>&lt;li>Shift from lines of code or features shipped to continuously improving quality and reducing the cost of change.&lt;/li>&lt;li>Track cognitive load through developer feedback on codebase clarity and ease of modification.&lt;/li>&lt;/ol>&lt;h3 id="collaboration">Collaboration&lt;/h3>&lt;ol>&lt;li>Promote cross-functional alignment by treating the team as the unit of work allocation.&lt;/li>&lt;li>Ensure accountability for high-quality implementations and bridge technical and business concerns.&lt;/li>&lt;/ol>&lt;h2 id="closing-thoughts-building-systems-that-last">Closing Thoughts: Building Systems That Last&lt;/h2>&lt;p>Taming complexity is about intentionality. By embedding these principles into your processes, you create codebases that support rapid iteration, scale gracefully, and don’t fall apart under continuous change. This isn’t about perfection but equipping teams to deliver value consistently without burning out or blowing budgets.&lt;/p>&lt;p>Look at your current projects: where’s complexity creeping in? Maybe it’s tangled dependencies or unclear module boundaries. Pick one area, apply a principle like explicit dependencies or test-first design, and observe the impact. What’s the worst complexity mess you’ve faced, and how did you tackle it? Share your experiences in the comments to keep this conversation alive.&lt;/p>&lt;p>Here’s to building software that evolves without fighting back!&lt;/p></description></item><item><title>'Winter hats go with winter coats', and 'supporters instead of jailers'.</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-31-supporters-not-jailers/</link><pubDate>Sun, 04 Feb 2024 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-31-supporters-not-jailers/</guid><description>&lt;p>In my journey through various workplaces, I&amp;rsquo;ve noticed recurring problems that seem to plague most companies. Two issues stand out prominently: the pervasive presence of office politics and the diminishing capability of companies over time.&lt;/p>&lt;p>But what if these issues were not just avoidable but had a remarkably simple solution that every company could implement?&lt;/p>&lt;h1 id="the-unavoidable-politics-conundrum">The Unavoidable Politics Conundrum&lt;/h1>&lt;p>Office politics, with its blame games and antagonistic dynamics, is often considered an inevitable part of corporate life. But is it truly unavoidable?&lt;/p>&lt;p>Imagine an ideal workplace where professionals collaborate seamlessly, work toward common goals, and collectively strive for improvement.&lt;/p>&lt;p>Does this sound like your workplace? Probably not. The reason for this disconnect lies in deeply ingrained beliefs that steer individuals toward a state of discord. As the saying goes, &amp;ldquo;most of the evil in this world is done by people with good intentions.&amp;rdquo; In the realm of corporations, good intentions manifest in the form of the separation of responsibility, and in the pursuit of control and resource efficiency.&lt;/p>&lt;h2 id="the-pitfalls-of-separation-of-responsibility">The Pitfalls of Separation of Responsibility&lt;/h2>&lt;p>Companies typically begin with the formation of an executive team, which then allocates global functions to various executives: Marketing, Revenue, Technology, Product, Operations, and so on. This separation cascades through the management hierarchy, creating sub-functions within each global function. For instance, within Technology, there are IT, engineering, Cloud infrastructure, security, and data sub-functions.&lt;/p>&lt;p>So what&amp;rsquo;s wrong with separating responsibility? It&amp;rsquo;s deeply anti-systemic, and it destroys alignment.&lt;/p>&lt;h2 id="the-lawyer-example">The Lawyer Example&lt;/h2>&lt;p>If you&amp;rsquo;re a lawyer, you belong to the Legal function, as your manager and their manager, all the way up to a C-level in Legal or Operations.&lt;/p>&lt;p>You sit and hang out with your peers, but you can&amp;rsquo;t seem to catch a break: everybody comes to you with all sorts of crazy ideas. You joke with your pals over drinks, laughing at how unthinkable some of these requests are:&lt;/p>&lt;ul>&lt;li>A team wants to launch a new product in China.&lt;/li>&lt;li>Somebody wants to fire a person from an ethnic minority background.&lt;/li>&lt;li>A board member is proposing to buy a company that has a development centre in Russia.&lt;/li>&lt;/ul>&lt;p>The risks are insane! Why does the company even hire these immature people? No way you&amp;rsquo;re going to let them put your company in danger. No way. Nope. Over your dead body.&lt;/p>&lt;h2 id="the-incentives-are-skewed">The Incentives are Skewed&lt;/h2>&lt;p>Human psychology plays a pivotal role here, as individuals inherently act in their self-interest. People do what they feel is best for themselves, and are amazing at instinctively knowing what that is, by unconsciously processing signals and clues in the environment they operate in.&lt;/p>&lt;p>At work, employees want to believe they&amp;rsquo;re doing a great job, and they don&amp;rsquo;t want to be fired. To orient themselves, they identify who their leader is, and try to keep them happy. This is not always their manager, but it&amp;rsquo;s always someone who can make or break the employee&amp;rsquo;s career at the company.&lt;/p>&lt;p>And what do you think happens when what&amp;rsquo;s best for the company differs from what&amp;rsquo;s best for an employee? This misalignment between personal goals and the overall objectives of the team or company leads to suboptimal decision-making.&lt;/p>&lt;p>It&amp;rsquo;s important to stress that our lawyer from the example above wants to do a great job. Do you see the problem? The issue is that our lawyer&amp;rsquo;s definition of a good job is not be aligned with what their company wants to achieve.&lt;/p>&lt;p>Our lawyer has everything to lose from considering any of the ideas other groups come to them with, and nothing to gain. If anything goes wrong, they can say goodbye to their bonus, or even lose their job. And they&amp;rsquo;re not getting a promotion if a new product line successfully launches in China.&lt;/p>&lt;p>Every time a company puts incentives in place, implicitly or explicitly, it drives the behavior of its employees.&lt;/p>&lt;h2 id="control-and-resource-efficiency-make-things-worse">Control and Resource Efficiency Make Things Worse&lt;/h2>&lt;p>By asking the global functions to enforce standard policies across the various business ventures, companies make things even worse. It&amp;rsquo;s as well-intentioned as it is naive. People that haven&amp;rsquo;t got a solid grasp of systems thinking inevitably fall into the trap of control and resource efficiency.&lt;/p>&lt;ul>&lt;li>&amp;ldquo;There&amp;rsquo;s not enough work for a DevOps engineer in each team&amp;rdquo;.&lt;/li>&lt;li>&amp;ldquo;You don&amp;rsquo;t want each initiative to have a different Cloud provider&amp;rdquo;.&lt;/li>&lt;li>&amp;ldquo;We&amp;rsquo;re a Java 8 house, and all our projects share a common Maven parent pom.xml to ensure we re-use what works&amp;rdquo;.&lt;/li>&lt;li>&amp;ldquo;The global security and architecture teams need to review and approve all changes&amp;rdquo;.&lt;/li>&lt;li>&amp;ldquo;Global IT ensures employees cannot just install whatever they want on their work laptops&amp;rdquo;.&lt;/li>&lt;/ul>&lt;p>Add dependencies between various siloed groups with misaligned incentives, and the result is chaos. Everything takes forever, people get frustrated and, over time, true inside wars develop. The first victim in this battlefield is the company itself, which becomes incapable of doing anything effectively.&lt;/p>&lt;p>After all, you can hire all the great people you want, but if they&amp;rsquo;re incapable and unwilling of working together towards the company&amp;rsquo;s objectives, nothing good will come out of it. A bad system beats a good person every time.&lt;/p>&lt;h1 id="the-solution-a-paradigm-shift-in-organizational-design">The Solution: A Paradigm Shift in Organizational Design&lt;/h1>&lt;p>So how do we avoid all of this? What are the secret organizational principles that magically align everybody? The remedy to these challenges lies in three fundamental organizational principles:&lt;/p>&lt;p>&lt;strong>1 - Initiative-Based Structure&lt;/strong>&lt;/p>&lt;p>Instead of organizing people by function, group them by initiative. Whether you&amp;rsquo;re a lawyer, a software developer, a designer, an HR person, or any other specialist, if you influence a business initiative, you report exclusively to the head of that business initiative, irrespective of their background. The initiative heads report directly into the founder or CEO.&lt;/p>&lt;p>&lt;strong>2 - Cross-Functional Equivalent Teams&lt;/strong>&lt;/p>&lt;p>Establish teams with diverse skill sets that can address any task within the initiative. Each team member, regardless of their primary expertise, reports to the same person, who&amp;rsquo;s also within the initiative.&lt;/p>&lt;p>&lt;strong>3 - Supporters instead of jailers&lt;/strong>&lt;/p>&lt;p>Contrary to concerns about standardization and control, embracing diversity within initiatives is crucial. Over time, companies deteriorate as new endeavors become stifled by rigid company-wide constraints. Allowing initiatives to operate differently fosters innovation and adaptability.&lt;/p>&lt;p>While it might seem counterintuitive, decentralizing support functions like IT, Legal, HR, and Finance for each initiative is essential. These functions should act as supporters, not enforcers. Shared resources are permissible, but their engagement and involvement should be at the discretion of the initiative, ensuring they remain aligned with the initiatives&amp;rsquo; goals.&lt;/p>&lt;p>Within an initiative, if some people work outside a team, then those people are advisors and supporters, not policy-makers. If you want to have a security function within a business initiative, that&amp;rsquo;s fine, but they advise and help the teams, without ever reviewing, approving, or enforcing anything.&lt;/p>&lt;h1 id="a-call-to-action">A Call to Action&lt;/h1>&lt;p>In the realm of organizational design, adopting a systems thinking perspective is transformative. By restructuring reporting lines and fostering cross-functional collaboration within initiatives, companies can break free from the shackles of politics and stagnation. The key lies in creating a fleet of nimble, adaptable units rather than a colossal vessel weighed down by internal conflicts.&lt;/p>&lt;p>As you embark on the journey of building or reshaping your company, remember the two fundamental rules: &amp;ldquo;winter hats go with winter coats&amp;rdquo; (Jason Gorman said this while giving advice about component-level architecture, but it&amp;rsquo;s universal) and &amp;ldquo;supporters instead of jailers&amp;rdquo;. These principles will not only enhance collaboration and productivity but will also create a workplace where individuals thrive and organizations evolve seamlessly.&lt;/p></description></item><item><title>A vision for Web 5</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-19-a-vision-for-web-5/</link><pubDate>Fri, 19 Jan 2024 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-19-a-vision-for-web-5/</guid><description>&lt;p>Hype cycles in tech rarely translate to significant changes in our lifestyle. Technology is a wide topic, but let&amp;rsquo;s focus on information systems today, and leave energy, food, water, health, and other important topics for another time.&lt;/p>&lt;p>We made it to 2024, and the current frenzy is about AI. Some people claim it&amp;rsquo;ll change everything. I&amp;rsquo;m quite skeptical, but let&amp;rsquo;s not digress, and think about the web instead.&lt;/p>&lt;p>If you think about it, the web hasn&amp;rsquo;t significantly changed in the past 20 years or so. Sure, things look better and work faster, but Google Maps came out in 2005. There was a lot of talk about Web 3 and all the changes it should have brought.&lt;/p>&lt;p>So let&amp;rsquo;s assume we finally got tired of NFTs, AI, unicorns, and mini-ponies with Web 4, and we somehow decided to actually try and improve things with Web 5. What would that look like?&lt;/p>&lt;p>In this post I try to imagine what the web could be like, to make people&amp;rsquo;s life easier. I feel we reached a point where further improvements in speed won&amp;rsquo;t significantly improve the experience, so the changes should qualitative.&lt;/p>&lt;h1 id="web-5-ideas">Web 5 ideas&lt;/h1>&lt;p>There are some requirements, for these ideas to work:&lt;/p>&lt;p>We should build around quantum-secure cryptography. Example algorithms could be Kyber as a key encapsulation mechanism, and Dilithium for signatures. The algorithms are just examples, as they&amp;rsquo;re not confirmed to be fully secure yet.&lt;/p>&lt;p>High-speed internet access should be available country-wide, for free. Other country-wide free services should include a checking account with payments, and a telecommunication service supporting audio and video.&lt;/p>&lt;p>It&amp;rsquo;s important that the government offers these, for free, because the assumption is that everybody should have access to these, in order for some of the points below to work.&lt;/p>&lt;p>Finally, the government should issue every citizen an identity, with a passive chip, falling back to a mobile device.&lt;/p>&lt;h2 id="governments-as-certificate-authorities-for-digital-identity">Governments as Certificate Authorities for digital identity&lt;/h2>&lt;p>Governments could act as Certificate Authorities, issuing quantum-secure certificates, associating a public key with the identity of a person or an organization.&lt;/p>&lt;p>People could verify the authenticity and validity of a certificate using a local cache of the public keys of various governments, without requiring internet access and a connection to a server.&lt;/p>&lt;p>Well-known government endpoints would allow people to receive the public key of a government.&lt;/p>&lt;p>Some of you might scream at the idea of a government responsible for identity management, and wish for a de-centralized approach. Not only things already work like this though, but a central trusted authority is needed to address cases where an identity is somehow lost. Imagine losing your chip or private key, suddenly the idea of a government able to issue you another key-pair associated to your identity sounds pretty great.&lt;/p>&lt;h2 id="trust-based-distributed-ledger">Trust-Based Distributed Ledger&lt;/h2>&lt;p>A Trust-Based Distributed Ledger (TBDLT) could help society leverage information in a trusted way. This would only be for information, not for currency. I&amp;rsquo;m not talking about Blockchain.&lt;/p>&lt;p>The idea is that an entity - a person or an organization - could attest a statement. Statements have a schema, and some data, and can be encoded as bytes, so that an attesting entity could sign them cryptographically.&lt;/p>&lt;p>Anybody could attest anything. The ledger would be decentralized, with local storage. Such a ledger would support storing, sharing, and requesting statements, along with the attestations and the identity of the attestors. Centralized services could provide convenience e.g., storing statements in the Cloud, to have them accessible across devices, and not to lose them.&lt;/p>&lt;p>About timestamps:&lt;/p>&lt;ol>&lt;li>Every statement includes a creation timestamp, and can optionally include an expiry timestamp to show the limit of its validity.&lt;/li>&lt;li>Timestamps are attested by notary services, run by governments and large corporations. The creator of the statement includes a timestamp, and then asks a notary to attest it. The notary only attests it if the difference between the proposed creation time and the current time is within a narrow margin.&lt;/li>&lt;/ol>&lt;p>An entity could delegate the ability to attest some statements to another entity. This is also a statement, so it can come with validity limits, etc.&lt;/p>&lt;p>Mind that here cryptography does not replace trust, but enhances it. A signature only proves that a person or an organization supports the validity of a statement. You&amp;rsquo;re never force to take anyone&amp;rsquo;s word for something though. It&amp;rsquo;s up to you to decide who to trust and why.&lt;/p>&lt;p>Users can customize their trust preferences, including:&lt;/p>&lt;ol>&lt;li>Requesting specific attestors - or attestor category - for a given statement type or category. An example is that you might decide to only trust the British Government to attest the right of a British citizen to travel internationally.&lt;/li>&lt;li>Requesting a minimum number of attestors.&lt;/li>&lt;li>Only considering attestors within a maximum degree of delegation distance.&lt;/li>&lt;/ol>&lt;p>When an entity presents a statement to another entity, they can include a set of supporting statements, and the receiving entity could use this information bundle to decide whether to trust the original statement being presented.&lt;/p>&lt;p>&lt;strong>Example: British citizenship&lt;/strong>&lt;/p>&lt;ul>&lt;li>I can decide to trust a statement about a person being a British Citizen only if anybody working for the British government at the time signed it.&lt;/li>&lt;li>The worker can convince me they were working for the British government at the time, by sharing a statement signed by other employees, all the way up to a key belonging to the government itself.&lt;/li>&lt;li>I can check the timestamps of all statements involved in providing trust against the public keys of the notaries who signed them.&lt;/li>&lt;li>These checks require no internet access.&lt;/li>&lt;/ul>&lt;p>&lt;strong>Example: train tickets&lt;/strong>&lt;/p>&lt;ul>&lt;li>I purchase a train ride, and receive a cryptographically-signed statement with details about my entitlement.&lt;/li>&lt;li>When I show up at the gate, the gate gets my entitlement through NFC, and verifies the signature against its own trust policies.&lt;/li>&lt;li>Since the train company signed my entitlement to get on board, and since the current time falls within the allowed time window confirmed against the notary&amp;rsquo;s signature, the gate opens.&lt;/li>&lt;li>These checks require no internet access.&lt;/li>&lt;/ul>&lt;h2 id="citizen-mobile-application">Citizen mobile application&lt;/h2>&lt;p>The government should issue a mobile application to all citizens, for free. This one covers:&lt;/p>&lt;ul>&lt;li>Checking account management.&lt;/li>&lt;li>Entity-to-entity payments.&lt;/li>&lt;li>Authentication manager.&lt;/li>&lt;li>Password manager.&lt;/li>&lt;li>Telecommunications e.g., voice and video calls.&lt;/li>&lt;li>The application for the Trust-Based Distributed Ledger.&lt;/li>&lt;/ul>&lt;h2 id="digital-identity-at-the-center">Digital identity at the center&lt;/h2>&lt;p>Digital identity should be at the center of all services and applications:&lt;/p>&lt;ul>&lt;li>It acts as username for all cases.&lt;/li>&lt;li>Cryptographic challenges can involve application-specific secrets (stored encrypted within the government mobile application).&lt;/li>&lt;/ul>&lt;p>This means no more:&lt;/p>&lt;ul>&lt;li>Email addresses.&lt;/li>&lt;li>Phone numbers.&lt;/li>&lt;li>Usernames for applications.&lt;/li>&lt;li>Physical addresses for shipping things (the address is requested using the TBDLT by the shipping company).&lt;/li>&lt;li>Passports (as the identity acts as target for entitlements that allow the person to travel internationally).&lt;/li>&lt;li>Bank account details (the identity is the target for peer-to-peer payments).&lt;/li>&lt;/ul>&lt;h2 id="cashless-society">Cashless society&lt;/h2>&lt;p>No more carrying around money, or having to use an ATM. This also means:&lt;/p>&lt;ul>&lt;li>No tax evasion (as all banks need to communicate all transactions to the central bank, so tax can be calculated directly).&lt;/li>&lt;li>A serious blow to crime, especially organized crime.&lt;/li>&lt;/ul>&lt;h2 id="job-specific-workflows-and-semantic-contracts">Job-specific workflows and semantic contracts&lt;/h2>&lt;p>The interactions entities have on the internet should be redesigned to be job-specific and declarative. Semantic contracts can ensure this is feasible in terms of implementation. These contracts:&lt;/p>&lt;ul>&lt;li>Describe what commands, queries, and subscriptions an entity allows, along with how to perform those invocations.&lt;/li>&lt;li>Use a versioned schema, evolved and published using the Trust-Based Distributed Ledger.&lt;/li>&lt;li>Completely abstract the medium used to communicate, along with any medium-specific piece of information e.g., a phone number.&lt;/li>&lt;/ul>&lt;p>Subscriptions allow people to communicate information changes to interested parties.&lt;/p>&lt;ul>&lt;li>They are pushed-based, sent by the party that changes their information.&lt;/li>&lt;li>Interested parties request to be notified of changes in specific pieces of information (statements), based on their type or category. Entities can accept or reject these requests.&lt;/li>&lt;li>Subscription requests can include an expiry timestamp.&lt;/li>&lt;li>When the information changes, the party that changed the information proceeds to notify the interested parties with peer-to-peer calls.&lt;/li>&lt;li>A centralized notification service can be used for convenience, falling back to decentralized peer-to-peer invocations if it&amp;rsquo;s unavailable.&lt;/li>&lt;/ul>&lt;p>Workflows can be composed based on different commands, queries, and subscriptions.&lt;/p>&lt;p>&lt;strong>Example: voice calls&lt;/strong>&lt;/p>&lt;ul>&lt;li>A person exposes commands to request being added to their inbound whitelist, and to request initiating a voice call.&lt;/li>&lt;li>The telecommunication service only allows to request to initiate a voice call with someone if the caller is registered to the inbound whitelist of the receiver.&lt;/li>&lt;li>When you want to be allowed in, you first have to request it with the person you want to call in the future. This request can come with an expiry timestamp.&lt;/li>&lt;li>If you&amp;rsquo;re allowed to call, you can request to initiate a voice call, and the service routes that to the person.&lt;/li>&lt;li>If the person accepts the call, you&amp;rsquo;re talking over voice.&lt;/li>&lt;li>Government services would come pre-enabled in the whitelist, so that they can request a call at any time.&lt;/li>&lt;li>The identity of both parties is verified with asymmetric cryptography.&lt;/li>&lt;li>You don&amp;rsquo;t need any protocol-level information e.g., a phone number. You simply request to speak with the entity.&lt;/li>&lt;li>This would eliminate most scams that happen over the phone, and all undesired advertisements and services.&lt;/li>&lt;li>This would effectively remove the need for landlines and the mobile network, provided that high-speed internet is publicly available country-wide.&lt;/li>&lt;/ul>&lt;p>&lt;strong>Example: shipping goods&lt;/strong>&lt;/p>&lt;ul>&lt;li>People can ship packages to an entity, without having to know their physical address.&lt;/li>&lt;li>Government shipping services can request to subscribe to changes in the physical address of an entity.&lt;/li>&lt;li>When an entity changes their physical address, the government shipping service is notified.&lt;/li>&lt;li>Shipped goods reach their target entity regardless of changes to their physical address.&lt;/li>&lt;/ul>&lt;p>&lt;strong>Example: booking a restaurant&lt;/strong>&lt;/p>&lt;ul>&lt;li>A restaurant exposes queries to see the menu, the opening times, and the free slots.&lt;/li>&lt;li>They also expose commands to make a reservation, and to get in contact over voice (if previously whitelisted).&lt;/li>&lt;li>A person browses the menu, checks when a free slot is, and requests to book a table at a time.&lt;/li>&lt;li>Given the semantic nature of these commands, they can do this with a few clicks, and without dealing with phone numbers or downloading PDFs.&lt;/li>&lt;/ul>&lt;p>&lt;strong>Example: getting prescription-based medicines&lt;/strong>&lt;/p>&lt;ul>&lt;li>A doctor issues and attests a statement acting as entitlement for a person to access a specific prescription-based medicine, with a specific dosage.&lt;/li>&lt;li>The person also receives statements supporting the right of the doctor to issue and attest such a statement, typically another statement plus attestation about their belonging to a doctors&amp;rsquo; society.&lt;/li>&lt;li>The person walks into a pharmacy, exposes the data through NFC, including their own identity and supporting attestations, and receives the medicines they need.&lt;/li>&lt;li>This works without access to the internet at any stage, and without the need to communicate codes or anything.&lt;/li>&lt;/ul>&lt;p>&lt;strong>Example: emergency escalation&lt;/strong>&lt;/p>&lt;ul>&lt;li>A person has access to a command that notifies subscribers that they&amp;rsquo;re in danger.&lt;/li>&lt;li>People (relatives), organizations (insurance companies, GPs), and government services (police, ambulance service) can subscribe to these, provided the person accepts.&lt;/li>&lt;li>When the person feels in danger, they invoke the command, and all the interested parties are notified.&lt;/li>&lt;li>The person in danger can pre-authorize certain interested parties to invoke certain commands and to get some (temporary) subscriptions accepted automatically.&lt;/li>&lt;li>So when the person declares that they&amp;rsquo;re in danger, some interested parties can be granted a subscription to the person&amp;rsquo;s position (through GPS), and know their heart rate and vitals.&lt;/li>&lt;li>Workflows can be composed on both sides, so on receiving notification that somebody is in danger, an entity could automatically trigger a response on their side e.g., requesting people closest to the location of the person in danger to intervene.&lt;/li>&lt;/ul>&lt;h1 id="final-words">Final words&lt;/h1>&lt;p>That&amp;rsquo;s a wrap folks. These are some of the idea I believe would really improve society in terms of information systems. As always, let me know what you think!&lt;/p></description></item><item><title>Interdependencies in your system of work, and how not to evaluate software practices</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-10-interdependencies-in-your-system-of-work/</link><pubDate>Wed, 10 Jan 2024 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2024/2024-01-10-interdependencies-in-your-system-of-work/</guid><description>&lt;p>I hear of people saying their team tried Trunk-Based Development, and decided they didn&amp;rsquo;t find it beneficial after a couple of months. I&amp;rsquo;m not going to war about the benefits of Trunk-Based Development today, even if I do love this practice.&lt;/p>&lt;p>Instead, I want to discuss the fact that this kind of statement is problematic. You see, Trunk-Based Development is a practice, and therefore it&amp;rsquo;s part of a system of work. The performance of a system depends on how the parts interact, not on how the parts perform separately.&lt;/p>&lt;p>So you cannot look at Trunk-Based Development, a practice, and assess its performance in isolation from its containing system. It makes no sense whatsoever.&lt;/p>&lt;p>Also, experiencing something without a containing system is impossible. So when people say they tried Trunk-Based Development, they probably attempted to work directly on the main branch, and this was the only change in how they were working before.&lt;/p>&lt;p>What are the chances that &amp;ldquo;improving&amp;rdquo; a part of a system improves the performance of that system? Pretty much zero. Things actually almost always get worse.&lt;/p>&lt;p>When you introduce any practice, Trunk-Based Development is just an example, to the way you work, you&amp;rsquo;re implicitly creating a new system of work. And that&amp;rsquo;s what you experience in terms of performance. Not the merits of the practice itself, but the merits of your overall new work system.&lt;/p>&lt;p>It&amp;rsquo;s as nonsensical as replacing the thrusters of an airplane with a helicopter&amp;rsquo;s rotor, witnessing the crippled thing crash, and concluding that rotors are a bad idea. You might even be tempted to swap the rotor on the helicopter with some thrusters.&lt;/p>&lt;p>Once again, what are the chances that swapping a part of a system with a part of another system significantly improves the performance? Zero, and not just for aircraft.&lt;/p>&lt;h2 id="comparing-end-to-end-alternative-systems">Comparing end-to-end alternative systems&lt;/h2>&lt;p>The way people work to develop a software system is their system of work. This is a subsystem of the product development system, which in turn is a subsystem of the company, which is contained in the industry, and the economy, and so on ad infinitum. For the sake of this discussion, though, let&amp;rsquo;s stop at the way people develop a software system.&lt;/p>&lt;p>The system of work is a sociotechnical system, involving people, teams, processes, practices, and the software system itself. All these aspects, like in any system, are interdependent: you cannot divide a system into independent subsystems.&lt;/p>&lt;p>So the practices, even when considered all together, don&amp;rsquo;t have an independent effect on the system of work. You cannot take Test-Driven Development, Continuous Delivery, and Trunk-Based Development, and separate them from how the code is written, and from how the teams work.&lt;/p>&lt;p>People must experience end-to-end alternative systems of work to meaningfully evaluate the merits of different approaches. It&amp;rsquo;s about comparing an airplane with a helicopter, not an airplane with the same airplane with its thrusters replaced by a rotor.&lt;/p>&lt;p>The effectiveness obviously depends on the context and the task at hand. You wouldn&amp;rsquo;t use an airplane where helicopters excel, nor the other way around. There are some circumstances, though, in which either will do, and these tend to be leisure cruises with no stakes or associated risks. In software development it&amp;rsquo;s the same: if what you&amp;rsquo;re doing is trivial, or if you don&amp;rsquo;t care about the outcomes, work in any way you want. No point in thinking about it, do whatever, and you&amp;rsquo;ll get whatever.&lt;/p>&lt;p>If you happen to care, though, the difference in performance between two alternative systems can be tremendous. In particular, a system designed holistically and intentionally will usually dramatically outperform a system that&amp;rsquo;s a patchwork of practises put together like if their effect added up linearly.&lt;/p>&lt;h2 id="desired-properties-for-our-system-of-work">Desired properties for our system of work&lt;/h2>&lt;p>The first thing to do when attempting to design a system is to be explicit about its desired properties. After all, how can you design anything if you don&amp;rsquo;t know what you want?&lt;/p>&lt;p>Since adopting specific practices is not a goal, what properties should our system of work have? If we&amp;rsquo;re talking about developing software systems, I&amp;rsquo;d go with:&lt;/p>&lt;ol>&lt;li>Effectiveness: doing the right thing, meeting the needs of the customers and of the company, delivering value earlier.&lt;/li>&lt;li>Safety: not breaking things for customers, not losing data, meeting security, privacy, compliance, and legal needs.&lt;/li>&lt;li>Sustainability: no heroics, no sprints, no borrowing from the future, and not ending up increasing the required costs or efforts over time.&lt;/li>&lt;li>Resilience: tolerating temporary absences, dealing with late changes in requirements, and working flexibly on anything needed.&lt;/li>&lt;li>Efficiency: minimizing no-value activities. An activity is deemed no-value if it doesn&amp;rsquo;t directly add value for customers and users. Some useful activities, like discussing approaches, training employees, etc. are still no-value activities in this sense, because if you were able to provide the same value to your users without doing them, it&amp;rsquo;d be better.&lt;/li>&lt;/ol>&lt;p>If these are the desired properties we want in a system of work, we should compare alternative systems by assessing how well they do with regard to these properties.&lt;/p>&lt;p>I don&amp;rsquo;t feel like criticizing various systems of work I don&amp;rsquo;t believe in, at least not today. Instead, I&amp;rsquo;ll outline how a system of work I like operates, and reason around the properties it yields. I&amp;rsquo;ll leave it to you to assess the system of work your company employs.&lt;/p>&lt;h2 id="a-particular-system-of-work">A particular system of work&lt;/h2>&lt;p>Keep in mind that the various practices and processes are interdependent, so resist the temptation to decide whether you like this or that aspect, and form a holistic judgment at the end instead.&lt;/p>&lt;p>Also, a system of work is always the child of a belief system. So if any point evokes a strong negative reaction, stop for a minute and try to understand which of your fundamental believes is in conflict with that aspect. You don&amp;rsquo;t need to agree, but understanding why you strongly disagree will benefit you.&lt;/p>&lt;p>The proposed system would work according to the following principles:&lt;/p>&lt;ol>&lt;li>A cross-functional leaderless &lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-31-so-you-think-you-work-in-a-team/">team (not merely a group of coworkers)&lt;/a> of five product developers, working as equals, without seniority roles or separate responsibilities. This includes a mix of:&lt;ul>&lt;li>Skills at web, infrastructure, back-end, and mobile.&lt;/li>&lt;li>Settlers, explorers, and town planners.&lt;/li>&lt;li>People with creativity, with a bias for action, and with long-term thinking.&lt;/li>&lt;li>Starters and finishers.&lt;/li>&lt;li>Skills at scripting, prototyping, and testing.&lt;/li>&lt;li>A mix of backgrounds and experiences, the more, the merrier.&lt;/li>&lt;/ul>&lt;/li>&lt;li>The team members work in an ensemble e.g., with &lt;a href="https://www.agilealliance.org/resources/experience-reports/mob-programming-agile2014/">Mob Programming&lt;/a>. This means they work synchronously together on the same task, rather than splitting the job and working in isolation.&lt;/li>&lt;li>The team members are either all office-based, sitting closely with each other, or all fully-remote, working on a team video call most of the day.&lt;/li>&lt;li>The distributed system employs an &lt;a href="https://en.wikipedia.org/wiki/Event-driven_architecture">Event-Driven Architecture&lt;/a>. &lt;a href="https://camunda.com/blog/2023/02/orchestration-vs-choreography/">Service choreography&lt;/a>, &lt;a href="https://cqrs.files.wordpress.com/2010/11/cqrs_documents.pdf">CQRS&lt;/a>, and &lt;a href="https://www.eventstore.com/event-sourcing">Event Sourcing&lt;/a>. Services are unaware of each other, share no dependencies, and only communicate using &lt;a href="https://medium.com/@chaojie.xiao/domain-driven-design-practice-domain-events-15b38f3c58fc">Domain Event&lt;/a>s with a company-wide schema registry.&lt;/li>&lt;li>The services are grouped using &lt;a href="https://martinfowler.com/bliki/BoundedContext.html">Bounded Context&lt;/a>s, and following how the business works. &lt;a href="https://www.infoq.com/articles/ddd-contextmapping/">Context Mapping&lt;/a> is used to visualize how the various contexts interact. (&lt;a href="https://learn.microsoft.com/en-us/archive/msdn-magazine/2009/february/best-practice-an-introduction-to-domain-driven-design">Domain-Driven Design&lt;/a>).&lt;/li>&lt;li>Each software service is structured according to &lt;a href="https://www.arhohuttunen.com/hexagonal-architecture/">Hexagonal Architecture&lt;/a>. Driving and driven adapters, application, domain.&lt;/li>&lt;li>The team favors those programming languages with a sophisticated type system. Examples might be TypeScript for frontend applications, Kotlin/Scala for backend, Golang with &lt;a href="https://www.pulumi.com/b/">Pulumi&lt;/a> for infrastructure.&lt;/li>&lt;li>&lt;a href="https://martinfowler.com/bliki/DDD_Aggregate.html">Aggregate Roots&lt;/a>, &lt;a href="https://blog.jannikwempe.com/domain-driven-design-entities-value-objects">Entities, and Value Objects&lt;/a> form a rich and type-safe domain model for each &lt;a href="https://martinfowler.com/bliki/BoundedContext.html">Bounded Context&lt;/a>, &lt;a href="https://martinfowler.com/bliki/UbiquitousLanguage.html">reflecting the language used by the business&lt;/a>. (&lt;a href="https://learn.microsoft.com/en-us/archive/msdn-magazine/2009/february/best-practice-an-introduction-to-domain-driven-design">Domain-Driven Design&lt;/a>).&lt;/li>&lt;li>Shared libraries provide implemented patterns, utilities, and a few thin &lt;a href="https://deviq.com/domain-driven-design/shared-kernel">Shared Kernel&lt;/a>s.&lt;/li>&lt;li>There&amp;rsquo;s a company-wide registry for Pulsar/Kafka topics and contracts e.g., &lt;a href="https://www.openapis.org/">OpenAPI&lt;/a>, &lt;a href="https://avro.apache.org/">Avro&lt;/a>, and &lt;a href="https://json-schema.org/">JSON Schema&lt;/a>.&lt;/li>&lt;li>There&amp;rsquo;s &lt;a href="https://en.wikipedia.org/wiki/Extreme_programming_practices#:~:text=Collective%20code%20ownership%20">Collective Code Ownership&lt;/a>: no team or person owns any area of the codebase or any part of the product.&lt;/li>&lt;li>The team pulls initiatives from a company-wide queue, working on one thing at a time, end-to-end, till completion. The team is in continuous collaboration with stakeholders and external experts. Initiatives are carved to take between 1 week and 6 weeks each.&lt;/li>&lt;li>Initiatives are minimally specified, as a summary that captures the problem or opportunity. All conversations with stakeholders and experts happen synchronously (either face-to-face or on video call). All discussions include all the relevant people, and happen at the latest responsible moment, right at the beginning of the work.&lt;/li>&lt;li>The team remains responsible for most decisions. Experts provide advice on demand, stakeholders clarify the problem or opportunity space, but the team is in charge of the solution space, and of the implementation.&lt;/li>&lt;li>The team uses &lt;a href="https://www.jpattonassociates.com/wp-content/uploads/2015/03/story_mapping.pdf">Story Mapping&lt;/a> to design and document the solution, and &lt;a href="https://en.wikipedia.org/wiki/Event_storming">Event Storming&lt;/a> to design the implementation.&lt;/li>&lt;li>The code is written using &lt;a href="https://www.codecademy.com/article/tdd-outside-in">Outside-In Test-Driven Development&lt;/a>, with clear contracts and example tests at module boundary&amp;rsquo;s level.&lt;/li>&lt;li>Each invocation (commands and queries) is processed along with a bundle of information about the invocation itself. The &lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-10-start-with-a-skeleton-not-to-die-a-zombie/#:~:text=Correlation-,Invocation%20context,-Every%20invocation%20should">Invocation Context&lt;/a> contains the user ID, the tenant ID, the ID of the token, etc. This information is logged, and it&amp;rsquo;s part of each published &lt;a href="https://medium.com/@chaojie.xiao/domain-driven-design-practice-domain-events-15b38f3c58fc">Domain Event&lt;/a>. It&amp;rsquo;s also used to &lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-10-start-with-a-skeleton-not-to-die-a-zombie/#:~:text=message%20correlation%20information.-,Idempotency,-The%20invocation%20context">implement&lt;/a> &lt;a href="https://betterprogramming.pub/architecting-distributed-systems-the-importance-of-idempotence-138722a6b88e">idempotency&lt;/a> for all operations.&lt;/li>&lt;li>The services use a build system capable of only rebuilding and retesting the modules that actually changed, using a deterministic non-cryptographic hash algorithm to determine whether a module has changed. An example for the &lt;a href="https://en.wikipedia.org/wiki/Java_virtual_machine">JVM&lt;/a> is how &lt;a href="https://gradle.org/">Gradle&lt;/a> uses the &lt;a href="https://www.baeldung.com/gradle-build-cache">Build Cache&lt;/a> to achieve this.&lt;/li>&lt;li>The code is made of well-defined modules with clear boundaries, responsibilities, and contracts. These can include &lt;a href="https://docs.gradle.org/current/userguide/multi_project_builds.html">Gradle modules&lt;/a>, &lt;a href="https://legacy.reactjs.org/docs/components-and-props.html">React components&lt;/a>, shared libraries, (micro)services, etc. Limiting how these modules interact and establishing rigorous contracts, with tests checking them, is also essential, because it allows knowing when to stop testing things because of a change. An &lt;a href="https://en.wikipedia.org/wiki/Event-driven_architecture">Event-Driven Architecture&lt;/a> with data contracts using &lt;a href="https://avro.apache.org/">Avro&lt;/a> really helps achieve this.&lt;/li>&lt;li>The build performs an array of tests and checks, including:&lt;ul>&lt;li>Application tests: they test the application layer against the stubbed driven adapters. These cover the application logic.&lt;/li>&lt;li>Integration tests: driving adapters against a stubbed application, checking their inbound contract e.g., OpenApi or Avro. Driven adapters against their target technology running in Docker through TestContainers.&lt;/li>&lt;li>In-process service tests: the whole service going from driving adapter to driven adapters and back, against the downstream technologies running in Docker through TestContainers.&lt;/li>&lt;li>Container-based service tests: the service itself running in Docker, using TestContainers, from driving adapter to driven adapters and back, against the downstream technologies in Docker, also through TestContainers.&lt;/li>&lt;li>Property-based tests: where inputs are randomized, and outputs are checked against properties in relation to inputs e.g., the inverse of a matrix multiplied by the original matrix should produce the identity matrix.&lt;/li>&lt;li>UI-driven tests: these tests check the correct rendering of UIs by exercising the UI against an in-memory implementation of the SDK that abstracts all communications between frontend and backend.&lt;/li>&lt;li>Compliance checks for data contracts e.g., OpenAPI, JSON, and Avro, against company-wide rules e.g., paths are dash-separated, no typos, no top-level array fields, etc.&lt;/li>&lt;li>Performance micro-benchmarks. These tests prevent regressions in performance-sensitive parts of a service, without taking a big amount of time to run.&lt;/li>&lt;li>Mutation tests. These tests inject mutations in the codebase before running the other tests, and count how many mutants go through the tests without failures. This is a way of assessing the effectiveness of your tests.&lt;/li>&lt;li>Static code analysis, with regard to security, cyclomatic complexity, code smells, and inefficiencies.&lt;/li>&lt;li>Dependency scanning to check dependencies and Docker images for security vulnerabilities.&lt;/li>&lt;/ul>&lt;/li>&lt;li>Performing all the tests and checks that are part of every build takes less than 3 minutes overall. This is only possible with great attention to modularity, a smart build system, and tests that can confirm whether some changed module still honors its contracts.&lt;/li>&lt;li>The team works directly on the main branch, with &lt;a href="https://medium.com/@mattia.battiston/why-i-love-trunk-based-development-641fcf0b94a0">Trunk-Based Development&lt;/a>, pushing a commit every time a new test passes, and after every small refactoring or tidying. This typically happens at least once every 10 minutes, for each team.&lt;/li>&lt;li>&lt;a href="https://continuousdelivery.com/">Continuous Delivery&lt;/a> is in place: every commit pushed upstream to the main branch is built, tested, and packaged, and each packaged new version of the code is either released directly, with automatic rollbacks based on error-rate monitoring, or deemed releasable at any moment, without any further checks needed.&lt;/li>&lt;li>Only one permanent environment exists: production. There are internal tenants and users to test things before rolling them out to customers, but this &lt;a href="https://copyconstruct.medium.com/testing-in-production-the-safe-way-18ca102d0ef1">testing happens in production&lt;/a>.&lt;/li>&lt;li>New behavior is guarded by &lt;a href="https://martinfowler.com/articles/feature-toggles.html">Feature Flags&lt;/a>, so that it&amp;rsquo;s only visible to internal test tenants, until explicitly enabled for external tenants. This allows releasing features in a coordinated way with marketing and sales.&lt;/li>&lt;li>&lt;a href="https://ephemeralenvironments.io/">Ephemeral environments&lt;/a> that are identical to production can be brought up on demand, in minutes. These are used to test large-scale changes that cannot be gradually rolled-out.&lt;/li>&lt;li>&lt;a href="https://newrelic.com/blog/how-to-relic/smoke-testing-with-synthetic-monitors">Smoke tests run continuously in production&lt;/a>, on internal test tenants and users, simulating the company&amp;rsquo;s business workflows end-to-end. Alerts are triggered if these tests fail.&lt;/li>&lt;li>&lt;a href="https://www.lambdatest.com/learning-hub/chaos-testing">Chaos engineering&lt;/a> simulates the random loss of arbitrary infrastructure areas, in production.&lt;/li>&lt;li>An external monitor is used to check whether the APIs and web applications of the company are reachable from outside the company&amp;rsquo;s infrastructure, and for &lt;a href="https://www.pagerduty.com/resources/learn/dns-monitoring/">DNS monitoring&lt;/a>. Alerts are triggered if these checks fail.&lt;/li>&lt;/ol>&lt;h2 id="properties-of-this-work-system">Properties of this work system&lt;/h2>&lt;p>In terms of the properties this system of work has, they include:&lt;/p>&lt;ol>&lt;li>Feedback happens early and often. From continuously working with stakeholders and experts, from running tests and checks locally, and from releasing tens of times a day. This delivers value earlier and prevents large amounts of rework due to misunderstood requirements.&lt;/li>&lt;li>Nobody works alone. Eight pairs of eyes review each character typed. This raises quality, prevents mistakes, and avoids rework delayed inspection would introduce. Problems are identified earlier, when fixing them is much cheaper.&lt;/li>&lt;li>Types, tests, checks, ways of working, data contracts, CQRS, event-driven workflows, and a pre-defined component-level architecture all heavily limit what&amp;rsquo;s possible. Fewer possibilities mean fewer things to keep in mind, which means fewer mistakes.&lt;/li>&lt;li>Releases are very safe, because the team takes many more much smaller steps. The effort required to fix the occasional issue is low.&lt;/li>&lt;li>The teams have end-to-end ownership of an initiative, working directly with stakeholders, customers, and experts. Everybody works on the entire system. There are no silos, or dependencies among teams. The work stays interesting.&lt;/li>&lt;li>The amount of work-in-progress is under control. Work is pull-based. Teams don&amp;rsquo;t work on more than they can chew. Nobody burns out. Things are finished before other things are started. The degree of parallelism equals the number of teams.&lt;/li>&lt;li>People work together within a team for a long time. Managers can coach these teams into becoming more effective over time.&lt;/li>&lt;li>Code quality is very high, and the architecture is modular and extensible. The cost of modifying the software system doesn&amp;rsquo;t continuously increase over time.&lt;/li>&lt;li>Knowledge spreads within a team, because of ensemble programming, and across teams, because there are no areas of the codebase owned by any team.&lt;/li>&lt;li>Each team can work on anything the company needs. Initiatives can be flexibly scheduled regardless of their domain area.&lt;/li>&lt;li>Every person has a farly good understanding of how the whole software system works, and of how the whole product works. This greatly reduces the number of mistakes, and makes supporting an on-call rotation easy.&lt;/li>&lt;li>The codebase is always working, at each commit, both locally and remotely. A team can change what they&amp;rsquo;re working on at any point without effort. Late changes in requirements are accommodated immediately with minimum disruption.&lt;/li>&lt;li>A ton of non-value activities are avoided: ceremonies, planning, fire-fighting, awaiting permission, splitting work across teams or people, release management, branch management, merge conflicts, code reviews, manual regression testing, manual deployments, security reviews, architecture reviews, design reviews, etc.&lt;/li>&lt;/ol>&lt;h2 id="final-words">Final words&lt;/h2>&lt;p>If you only take one thing from this post, let it be this: you cannot evaluate a practice in isolation. You should always assess alternative end-to-end systems of work, because the effectiveness depends on how the various parts work together. So next time you encounter a practice you&amp;rsquo;re curious about, don&amp;rsquo;t try to introduce it in your existing way of working. Instead, try to understand what it enables, and to think at how you could build a different way of working, given this new possibility.&lt;/p>&lt;p>And if you have place for one more takeaway, a work system that leverages Test-Driven Development, Domain-Driven Design, Continuous Delivery, Trunk-Based Development, Ensemble Programming, small long-lived teams, Collective Code Ownership, Event-Driven Architecture, and careful modularity, can be incredibly effective. Compare that with how you work where you are, and make your own conclusions.&lt;/p>&lt;p>I hope this was useful to you and, as always, let me know what you think, especially if you disagree.&lt;/p></description></item><item><title>So you think you work in a team</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-31-so-you-think-you-work-in-a-team/</link><pubDate>Sun, 31 Dec 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-31-so-you-think-you-work-in-a-team/</guid><description>&lt;p>I had a discussion on Twitter yesterday about whether a software development team should restrict the freedom of its members, and enforce shared approaches and ways of working. I and others were adamant that yes, this should absolutely be the case. Others were horrified, saying this is everything that&amp;rsquo;s wrong with our industry.&lt;/p>&lt;p>This made me realize that I never wrote about teamwork in software product development. This is borderline insane, as I&amp;rsquo;m hugely passionate about system thinking and teams (which are systems, after all), and I have very strong opinions about this.&lt;/p>&lt;p>And so, here we are, let&amp;rsquo;s talk about teams, teamwork, and the role of management in software product development.&lt;/p>&lt;h2 id="you-probably-dont-work-in-a-team">You probably don&amp;rsquo;t work in a team&lt;/h2>&lt;p>Ask any software developer whether they work as a team, and sure as shooting they&amp;rsquo;ll tell you they do. But do they? I worked in quite a few places and, in my experience, people rarely work as teams in software companies.&lt;/p>&lt;p>Teamwork is often misunderstood. I guess the word &amp;ldquo;team&amp;rdquo; feels more personal, and evokes better emotions, and that&amp;rsquo;s why most people use it interchangeably with &amp;ldquo;group of co-workers&amp;rdquo;.&lt;/p>&lt;p>But a team is not simply a group of people working towards the same goal. And I believe that many of the problems companies experience derive from this misunderstanding, resulting in anti-systemic practices that prevent true teamwork.&lt;/p>&lt;h2 id="definition-of-a-team">Definition of a team&lt;/h2>&lt;p>So what&amp;rsquo;s a team then? A team is a social system, and as such it is a qualitatively different entity than a few individuals working alone side-by-side. To understand what this means, let&amp;rsquo;s look at what &amp;ldquo;system&amp;rdquo; and &amp;ldquo;social&amp;rdquo; mean, in this context.&lt;/p>&lt;p>When it comes to the definition of a system, I could never explain it better than by quoting Russell Ackoff:&lt;/p>&lt;ul>&lt;li>&amp;ldquo;A system is a whole that consists of parts each of which can affect its behavior or its properties. Each part of the system when it affects the system is dependent for its effect on some other part. In other words: The parts are interdependent.&amp;rdquo;&lt;/li>&lt;li>&amp;ldquo;No part of the system or a collection of parts of the system has an independent effect on it. Therefore, a system as a whole cannot be divided into independent parts.&amp;rdquo;&lt;/li>&lt;li>&amp;ldquo;The essential or defining properties of any system are properties of the whole which none of its parts have. The essential property of an automobile is that it can carry you from one place to another. No part of an automobile can do that.&amp;rdquo;&lt;/li>&lt;li>&amp;ldquo;Therefore, when a system is taken apart it loses its essential properties. A system is not the sum of the behavior of its parts, it’s a product of their interactions. The performance of a system depends on how the parts fit, not how they act taken separately.&amp;rdquo;&lt;/li>&lt;/ul>&lt;p>About the word &amp;ldquo;social&amp;rdquo;, Russell Ackoff, once again, categorized any system as either:&lt;/p>&lt;ul>&lt;li>A machine, where the overall system has no purpose, and neither have the parts. A machine serves the purpose of its creator.&lt;/li>&lt;li>An organism, where the overall system has its own purpose, but the parts do not. You have a purpose, but your liver, your brain, and your hands don&amp;rsquo;t.&lt;/li>&lt;li>A social system, where the overall system has its own purpose, and the subsystems also have their own purposes.&lt;/li>&lt;/ul>&lt;p>Therefore, a social system is a system whose parts have their own purpose. The beliefs and behaviors in a team are interdependent, and this interdependence develops and evolves over time, in a reciprocal manner.&lt;/p>&lt;p>Most of the issues and the dysfunction companies experience derive from either:&lt;/p>&lt;ol>&lt;li>Dealing with companies or teams in an anti-systemic way, by focusing on the parts, rather than on the system as a whole.&lt;/li>&lt;li>Treating companies, teams, and people as if they were machines, forgetting that they have their own purposes.&lt;/li>&lt;/ol>&lt;p>Today though, I want to focus on how most &amp;ldquo;teams&amp;rdquo; within software companies aren&amp;rsquo;t working as a system. I might come back another day to talk about the other mistakes mentioned above.&lt;/p>&lt;h2 id="the-difference-matters-a-lot">The difference matters a lot&lt;/h2>&lt;p>The reason I obsess about teams is that the difference in effectiveness between a team and a group of co-workers can be dramatic, in the right context.&lt;/p>&lt;p>Not all groups of co-workers are teams. As an example, street cleaners don&amp;rsquo;t typically work as a team. They all together have breakfast, get to the location, exchange jokes and banter, plan the work, and commute back. But behaving socially and having a common goal aren&amp;rsquo;t sufficient for a group to be called a team.&lt;/p>&lt;p>The defining characteristic of a team is that the team members work synchronously together toward the goal, without splitting the task and working individually.&lt;/p>&lt;p>Street cleaners divide the work among themselves, and then work individually. This works because cleaning streets is a linear activity, so if I clean a street, and you clean another, the overall outcomes are the sum of our individual outcomes. In these circumstances teamwork isn&amp;rsquo;t necessary, and a working group might even be better.&lt;/p>&lt;p>Working as a group or as a team is a choice, even if the outcomes are typically very different.&lt;/p>&lt;p>A good example is hunting. You can decide to hunt as a group, or as a team. As a group, you split the work, so I hunt in an area, and you in another. At the end of the hunt, we put together what you and I managed to get, and bring it back to the village. We work in parallel, more efficiently, and cover more ground. What&amp;rsquo;s not to love then?&lt;/p>&lt;p>Well, as long as we&amp;rsquo;re hunting things that are trivial to hunt, like frogs, this actually works best. What if we&amp;rsquo;re hunting something bigger or faster than us though? Perhaps in a dangerous environment with many dangerous predators? Humanity discovered a long time ago the value of working as a team, and that hunting alone was a great way of dying.&lt;/p>&lt;p>Remember? &amp;ldquo;The essential or defining properties of any system are properties of the whole which none of its parts have&amp;rdquo;. A team is capable of things none of its members could accomplish individually. This matters a lot when an activity is not linear, cannot be divided in independent activities, and the overall result is not the sum of the result of the smaller tasks.&lt;/p>&lt;p>We cannot hunt a part of a bear each. We&amp;rsquo;re either capable of hunting a bear as a team, or we cannot hunt one.&lt;/p>&lt;p>Only teams allow specialization of roles. If we&amp;rsquo;re splitting the work, you&amp;rsquo;re forced to be self-sufficient, because you&amp;rsquo;ll be working alone a lot. Teams, on the other hands, thrive on diversity. Adding more people that don&amp;rsquo;t bring new strengths doesn&amp;rsquo;t improve a team.&lt;/p>&lt;p>In Ocean&amp;rsquo;s Eleven, none of the gang&amp;rsquo;s members could have pulled the heist on their own. It&amp;rsquo;s only by acting as a team that something nobody could achieve became feasible. A gang made of only one type of specialist would have failed spectacularly.&lt;/p>&lt;p>Examples of teamwork are everywhere, not only in movies. A football team made of only goalkeepers, defenders, midfielders, or forwards wouldn&amp;rsquo;t go far. Even more significantly, a football team without any one of these roles also wouldn&amp;rsquo;t do well.&lt;/p>&lt;p>Once again, it&amp;rsquo;s the nature of the activity that determines the best approach. If an activity is not linear, splitting the work never makes sense. No competitive activity is linear, hence the importance and prevalence of teams in sports and war.&lt;/p>&lt;h2 id="software-development-is-definitely-not-linear">Software development is definitely not linear&lt;/h2>&lt;p>So the million-dollar question is whether software development is a linear activity. If it is, all good, and people can work how they want. But if it&amp;rsquo;s not a linear activity, groups of co-workers are going to be dramatically outperformed by true teams.&lt;/p>&lt;p>Software development isn&amp;rsquo;t a competitive activity, no other software team is there trying to outmaneuver and sabotage you, so you don&amp;rsquo;t need to adapt how you work to what your competitors are doing. Even in product development competition is more a fantasy some folks have than an actual threat to a company. Sure, other companies also want to capture the market, but you can decide not to compete, and to find your own niche other companies aren&amp;rsquo;t interested in. The goal is making a profit, not world domination. It&amp;rsquo;s not a zero-sum game.&lt;/p>&lt;p>Is there hope then? Hardly so. Software development can never be linear, because it&amp;rsquo;s about developing software systems. A system&amp;rsquo;s performance is never the sum of its parts, but the product of their interactions. So you can never split the development of a software system, work in parallel, put the pieces together, and end up with great outcomes.&lt;/p>&lt;p>I can hear some of you screaming. And indeed, there are some exceptions to what I just claimed. If a software system is trivial and tiny e.g., a low-throughput CRUD application with small amounts of data, people can easily divide the tasks and work alone in parallel. It might even be better.&lt;/p>&lt;p>In many cases, though, software systems are large and complex, having to support a large concurrent number of users while maintaining a low latency. They can also involve large amounts of data, have strict data integrity and privacy requirements, and must continue to work in presence of network partitions and hardware faults.&lt;/p>&lt;p>For such projects, individuals working in parallel cannot cope. That&amp;rsquo;s because, in an effective team, the individual strengths merge to form a collective understanding much higher than the sum of the parts. And so no team member could either design or build such a large system on their own, regardless of the time they have available, and yet the team can.&lt;/p>&lt;p>Product development is even less linear, since the goal is to find and address a need customers are willing to pay the company for. It&amp;rsquo;s not about building stuff, but about discovering problems, opportunities, and solutions. You cannot solve 1/6 of the puzzle and put all the pieces together.&lt;/p>&lt;h2 id="what-if-its-not-linear-after-all">What if it&amp;rsquo;s not linear after all?&lt;/h2>&lt;p>So what then? What changes if you believe that software product development is not a linear activity? As it turns out, you behave quite a few things.&lt;/p>&lt;p>First of all, you give up the fantasy that people are interchangeable. Unlike mechanical parts, two people are never identical. So you can never replace a team member without the whole team changing. Even worse, the same person or even the same pair of people will produce a different effect if they join different teams. A person that stabilizes a team might destabilize another. When discussing deliverables and initiatives, you now have to think about which people you need, rather than just how many.&lt;/p>&lt;p>Secondly, you end up using teams as the unit of work allocation. Assigning tasks to the individual team members makes no sense, because the outcomes are not the sum of the individual accomplishments. &amp;ldquo;Bob&amp;rsquo;s job will be hunting 1/6 of the bear&amp;rdquo;.&lt;/p>&lt;p>You only now hire people with a specific team in mind. Every team needs to be carefully assembled by mixing in various strengths, personalities, backgrounds, attitudes, and skills. So how could you ever look for generic &amp;ldquo;back-end engineers&amp;rdquo; company-wide, and then figure out where to place them? Unthinkable.&lt;/p>&lt;p>And the thought of assessing individual performance now makes you laugh. Somebody&amp;rsquo;s sense of humor or curiosity might contribute to the effectiveness of the team in ways that a more skilled and hardworking replacement could never hope to match.&lt;/p>&lt;p>Finally, the way you structure the incentives also changes. How can you reward a forward for every goal they score, knowing that winning the game depends on how well the team plays together? What do individual incentives do to teamwork? Wouldn&amp;rsquo;t that player be tempted to take every shot, rather than passing the ball? It might result in a higher bonus for themselves, after all, even if the team might win a lot less because of this.&lt;/p>&lt;h2 id="individual-freedom-teamwork-and-management">Individual freedom, teamwork, and management&lt;/h2>&lt;p>Let&amp;rsquo;s go back at the original point of contention on the Twitter discussion. A team must restrict the freedom of its members. This is a consequence of its systemic nature. The way a team works is designed holistically to produce the desired outcomes, and this is interdependent with the team members that end up making the team.&lt;/p>&lt;p>It should be obvious and a no-brainer. If the performance of the team is the product of the interactions between its members, how well you fit the team&amp;rsquo;s chosen approach matters way more than your individual skills. And I say chosen, because different teams can play in radically different ways, and achieve their goals.&lt;/p>&lt;p>The most important job of any manager is developing effective approaches, and then carefully assembling teams that are well-formed to play that way. The parallel with the world of sports and military is clear.&lt;/p>&lt;p>Within a team there&amp;rsquo;s no space for uncoordinated activity. You can be a great football player but, if you don&amp;rsquo;t want to pass the ball a lot, Barcelona is not the right team for you. You can be a great shot but, if you love whistling, and you do that while we&amp;rsquo;re sneaking our way through an enemy base, you&amp;rsquo;re a liability.&lt;/p>&lt;h2 id="parting-words">Parting words&lt;/h2>&lt;p>Whew, my last post for this 2023! I hope you now have an appreciation of the difference between a group of co-workers and a team, and that you can tell which one your company adopts.&lt;/p>&lt;p>I very vocally believe that software product development is a nonlinear activity best done in teams, and I strongly favor ensemble programming because of this. So if you want to work alone or remotely, be upfront and find a team that operates that way. But don&amp;rsquo;t fool yourself thinking this should be your choice after joining any team.&lt;/p>&lt;p>You don&amp;rsquo;t need to agree with me, but please let me know your thoughts, especially if you disagree!&lt;/p></description></item><item><title>My belief system for software product development</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-21-my-belief-system-for-software-product-development/</link><pubDate>Sun, 24 Dec 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-21-my-belief-system-for-software-product-development/</guid><description>&lt;p>Running a software product development company is complex. People, processes, products, technologies, and operations are all crucially important, and interdependent.&lt;/p>&lt;p>You need to identify a job people struggle with, create a product that addresses it well, price and market it appropriately, defend it against competitors, and react to market changes, all while making a profit.&lt;/p>&lt;p>Additionally, software is a highly abstract medium, meaning that common-sense rarely applies, so that what&amp;rsquo;s effective is often counter-intuitive.&lt;/p>&lt;p>So it&amp;rsquo;s perhaps unsurprising that software companies rarely know how to structure themselves and how to operate, leading to many alternative approaches. Startups copy practices from famous businesses, and there&amp;rsquo;s an abundance of consultants and advisors. Commercial frameworks like SAFe or Scrum promise executives that they won&amp;rsquo;t have to think, and that merely applying the framework will produce great results.&lt;/p>&lt;p>And yet, reality isn&amp;rsquo;t linear or clear-cut. What works in a company won&amp;rsquo;t likely work elsewhere. A practice that&amp;rsquo;s effective in a company with 30 employees will almost surely not work when the same company has 100 employees. Systems are full of interdependencies and nonlinearities, so you cannot improve your sales without changing your product and your marketing, and you cannot change who you hire without changing how you work.&lt;/p>&lt;p>So what should a company do when common sense doesn&amp;rsquo;t work, frameworks aren&amp;rsquo;t effective, and copying what famous businesses are doing won&amp;rsquo;t take you far?&lt;/p>&lt;p>The answer is that you should think deep and write down your belief system, and then use it to orient yourself in this highly complex environment. A belief system is a set of fundamental beliefs that shapes the way you perceive reality and links causes to consequences.&lt;/p>&lt;p>The various beliefs that are part of your belief system shouldn&amp;rsquo;t conflict. If your beliefs are in conflict, question your current beliefs, until the conflict disappears.&lt;/p>&lt;p>What follows are some of my beliefs, when it comes to software product development. When I say &amp;ldquo;my belief&amp;rdquo;, I mean that I stand by them, not that I came up with them myself. They served me well over the years, but I keep evolving them, and you shouldn&amp;rsquo;t take them at face value.&lt;/p>&lt;p>The important thing is to:&lt;/p>&lt;ol>&lt;li>Write down what you currently believe, rather than what you&amp;rsquo;d like to believe.&lt;/li>&lt;li>Assess the coherency of your belief system.&lt;/li>&lt;li>Check each practice, process, and approach you use or are considering using against your beliefs.&lt;/li>&lt;li>Reassess your belief system every time you experience something that your belief system struggles to explain. Don&amp;rsquo;t discount reality if it clashes with your mental model. Discount your mental model if it clashes with reality.&lt;/li>&lt;/ol>&lt;h1 id="holistic">Holistic&lt;/h1>&lt;ol>&lt;li>Products, people, technologies, processes, incentives, and company structure are interdependent. You shouldn&amp;rsquo;t design them independently.&lt;/li>&lt;li>We&amp;rsquo;re all going to be wrong a lot. Remember you can still be wrong, even when you&amp;rsquo;re dead sure you&amp;rsquo;re right. Keep an open mind.&lt;/li>&lt;li>Everything is possible, unless it violates the laws of physics. Some things can be prohibitively expensive, or undesirable.&lt;/li>&lt;li>When designing something, start from the desired outcomes, and walk back to the technologies and the approaches. Don&amp;rsquo;t decide whether to turn left or right without knowing how that decision fits with your travel plan.&lt;/li>&lt;li>Your fight is against entropy. Do nothing, and things will get worse. Order and desired states require a lot of active effort. Chaos and disorder are the natural state of things.&lt;/li>&lt;li>The most important job for an executive is to form a core group of high performers, and to remove people that would endanger this.&lt;/li>&lt;li>Growth is important, but development equally so. Growth is about doing more, in terms of users, revenue, etc. Development is about becoming better at the activities involved in operating the company.&lt;/li>&lt;li>Trust and verify. Treat people like adults. Allow people to do mistakes. Get rid of people who abuse your trust or who keep making the same mistakes. Do not restrict everybody&amp;rsquo;s freedom because a few people abused that freedom.&lt;/li>&lt;li>You cannot inspect quality into a product or a system. You need to build quality in, throughout the lifecycle.&lt;/li>&lt;li>At some point, quantity becomes quality. Many small issues eventually give your customers the impression of a poor product.&lt;/li>&lt;li>If you have formal authority over other people, how good you are at your job can be assessed by how infrequently you have to resort to formal authority.&lt;/li>&lt;li>Nobody cares about your business and its customers. Everybody&amp;rsquo;s in for themselves. The best people you can hire are those who are in for the fun of solving problems together. Align the incentives so that what&amp;rsquo;s best for the company is what&amp;rsquo;s best for the employee.&lt;/li>&lt;li>Resist the lure of metrics and KPIs. Indicators are useful, but should never become targets. Running a company without metrics and KPIs can be hard. Running a company primarily by metrics and KPIs is usually a disaster. Many important things are not measurable.&lt;/li>&lt;/ol>&lt;h1 id="processes">Processes&lt;/h1>&lt;ol>&lt;li>Bureaucracy kills. Stay lean, keep things flat and collaborative, ensure that every requirement comes with the name of the person behind it. Needs should be met, but the suggested ways of meeting those needs does not need to be observed.&lt;/li>&lt;li>The job of management is to ensure that your system of work is effective. This requires an intimate understanding of both how the work works, and of how it should work. Management by objectives is a travesty. Replace it with management by means.&lt;/li>&lt;li>Executives, managers, and product developers have different jobs. Use clear boundaries for decision-making. Avoid splitting your company into people who think and people who do.&lt;/li>&lt;li>Push authority information, rather than information to authority.&lt;/li>&lt;li>The teams of product developers should make most decisions: about the product, and about their ways of working.&lt;/li>&lt;li>Whenever people disagree, a prototype or test is the right way forward. Whenever people agree, still, a prototype or test prevents embarrassing mistakes.&lt;/li>&lt;li>The coordination effort required in a team, department, or company increases super-linearly with its number of people.&lt;/li>&lt;li>Early feedback prevents waste.&lt;/li>&lt;li>The unit of work allocation should always be a team.&lt;/li>&lt;li>Teams should abstract who&amp;rsquo;s working on what, and be resilient to holidays, illness, etc.&lt;/li>&lt;li>A team should be able to work on anything the company needs.&lt;/li>&lt;li>Experts working outside a team should be advisors. Advisors are not decision-makers. The authority should remain with the team getting the advice. Including architecture, security, product management, design, etc.&lt;/li>&lt;li>When people capable of doing the thing disagree with people who wouldn&amp;rsquo;t be capable themselves, the former should have authority.&lt;/li>&lt;/ol>&lt;h1 id="product">Product&lt;/h1>&lt;ol>&lt;li>Customers hire your product to do a job they would otherwise attempt to do in other ways. Nobody cares about your product, people care about the job your product does for them.&lt;/li>&lt;li>Product development is about finding a good job to be done, and tailoring your product to meet the customers&amp;rsquo; needs well, given the circumstances your users find themselves in.&lt;/li>&lt;li>A software product is a set of promises, not a set of features. A product is still the same product if the features evolve, but not if the needs it meets changes.&lt;/li>&lt;li>The biggest risk for a new company is solving a problem not worth solving.&lt;/li>&lt;li>Behind every bad idea that were people thinking it was an excellent idea. Write down your assumptions, and test these assumptions, before fully committing. Finding evidence should rarely require building things.&lt;/li>&lt;li>Being a big fish in a small pond is better than being a small fish in a big pond. The right starting pond can be expanded after you outgrow it.&lt;/li>&lt;li>People make emotional rather than rational decisions. Their perception of your company, their job, your product, and the competition will determine their behavior. Positioning is the battle for the mind of your prospects.&lt;/li>&lt;li>The number of features and bugs is irrelevant. One great feature might make your product. One bad bug can kill your company. Most features and bugs won&amp;rsquo;t move the needle.&lt;/li>&lt;/ol>&lt;h1 id="technology">Technology&lt;/h1>&lt;ol>&lt;li>A software system is not the sum of its parts: every time there&amp;rsquo;s a new requirement, you need to modify the software to meet this new requirement plus all the previous ones.&lt;/li>&lt;li>You cannot create an effective system by building its parts sequentially.&lt;/li>&lt;li>Consistency of technologies is overrated. Even if you&amp;rsquo;re right when you choose something, changing your mind when needed will require convincing everybody and changing things company-wide.&lt;/li>&lt;li>Consistency in architecture is underrated. You won&amp;rsquo;t benefit from any desirable system-wide properties, unless you design your whole system to achieve these.&lt;/li>&lt;li>Humans can only keep so many things in their head. Well-designed modules compartmentalize complexity.&lt;/li>&lt;li>The cost (effort, risk) involved in modifying a software system depends on its internal quality, which is how well the code is structured.&lt;/li>&lt;li>The cost (effort, risk) involved in modifying software depends on how easy it is to understand where to operate the change, and on how contained the change is in terms of testing.&lt;/li>&lt;li>The cost of modifying software because of new requirements always ends up trumping the initial cost involved in writing it.&lt;/li>&lt;li>The cost (effort, risk) involved in modifying software depends super-linearly on how many possibilities you need to take into account.&lt;/li>&lt;li>A strongly typed domain model makes undesired possibilities impossible.&lt;/li>&lt;li>Software architecture is about limiting the degrees of freedom involved in moving and storing information.&lt;/li>&lt;li>Storing the history of changes, as opposed to storing the state itself, allows amazing options.&lt;/li>&lt;/ol>&lt;h1 id="people">People&lt;/h1>&lt;ol>&lt;li>Staffing the best people you can find is the single most important thing for any company.&lt;/li>&lt;li>Every employee should want to do their best, and to continuously improve their best.&lt;/li>&lt;li>Team members should vary in strengths, backgrounds, and experience, but should have compatible belief systems.&lt;/li>&lt;li>The tell-sign of a high-performing team is the pride they take in their craftsmanship.&lt;/li>&lt;li>People should be passionate and have strong opinions. Bad opinions can be fixed, but a lack of passion cannot.&lt;/li>&lt;li>Teams should be co-located: either everybody sitting closely together most days, or everyone entirely remote.&lt;/li>&lt;li>The 10x developer is a myth. When you factor in how bad decisions compound over time, a 10x factor isn&amp;rsquo;t even near the difference between bad and great.&lt;/li>&lt;li>Fewer and better people will run circles around many more worse employees.&lt;/li>&lt;/ol></description></item><item><title>Start with a skeleton not to die a zombie</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-10-start-with-a-skeleton-not-to-die-a-zombie/</link><pubDate>Sun, 10 Dec 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-12-10-start-with-a-skeleton-not-to-die-a-zombie/</guid><description>&lt;p>Most companies die a zombie. They start lean and nimble, ship feature after feature, and delight their customers. Then, after eighteen to twenty-four months, the bubble pops, and suddenly they cannot do anything. They limp around painfully until they bite the bullet and re-write their systems, or kick the bucket.&lt;/p>&lt;p>So what happens? What&amp;rsquo;s this hidden killer that shows itself only after a year and a half or two? The problem is that most companies never plan their systems. They start with some functionality, get it out, check what the customers want, and keep doing this.&lt;/p>&lt;p>And it&amp;rsquo;s absolutely the right thing to do. With a big but, though. As a company, you can survive without a well-planned system in place for a while, but you&amp;rsquo;re living on borrowed time. The cost of changing your system steadily increases in this case, until after a year and a half or two, you guessed it, changing anything has a prohibitive cost.&lt;/p>&lt;p>When I say cost, by the way, I mean risk. So you could have a lot of money, but it wouldn&amp;rsquo;t help much. Adding people makes things significantly worse very quickly, as they come up with workaround after workaround.&lt;/p>&lt;p>So what options do you have? Well, it depends on how many resources you have in the beginning.&lt;/p>&lt;p>Most companies won&amp;rsquo;t have almost anything at the start, so make assumptions, validate them, and build prototypes until you identify your product market fit. At some point, you&amp;rsquo;ll have to build a new system from scratch, and migrate your customers to it. Don&amp;rsquo;t make the mistake of avoiding this step.&lt;/p>&lt;p>In other circumstances, you might start as a new large program or company within a parent company with a large amount of resources. In this case you know you&amp;rsquo;ll end up building some product, even if you don&amp;rsquo;t exactly know what, but you&amp;rsquo;re not overly concerned of running out of money. Have two separate systems, one that&amp;rsquo;s a throwaway prototype for quick experimentation, and another one that&amp;rsquo;s well-planned and extensible.&lt;/p>&lt;p>In any case, you cannot build a complex well-functioning system part after part, and you cannot retrofit a large system that&amp;rsquo;s not effective into one that works well. You&amp;rsquo;ll want to start from a bare-bone system that does very little, but does it well: a walking skeleton.&lt;/p>&lt;p>A walking skeleton it&amp;rsquo;s like a Minimum Viable Product, but for your software systems. Its purpose is to prove that your choices and trade-offs achieve the desired system-wide properties you have in mind.&lt;/p>&lt;p>There are a few differences across B2B and B2C, but it&amp;rsquo;s the technical ambitions that determine the scope of a walking skeleton. That&amp;rsquo;s its beauty: it has very little to do with your actual product, or even industry. So you could start building one in a way that&amp;rsquo;s orthogonal to your product discovery efforts.&lt;/p>&lt;p>It&amp;rsquo;s called &amp;ldquo;walking&amp;rdquo; because the software is runnable, and it exemplifies the fundamental architectural patterns, approaches, technologies, and moving parts. And it&amp;rsquo;s called &amp;ldquo;skeleton&amp;rdquo; because it&amp;rsquo;s a robust system that works, but it does nothing significant: you&amp;rsquo;ll add your product-specific behavior to it after it&amp;rsquo;s ready.&lt;/p>&lt;p>Starting your software system from a walking skeleton offers several benefits:&lt;/p>&lt;ol>&lt;li>Designing the various parts and aspects together improves the quality of the overall system. This is because the parts are inter-dependent, so you cannot design one aspect without keeping in mind all the others. Simply keeping in mind all the aspects will force you to design things differently.&lt;/li>&lt;li>It yields a software system that works end-to-end, so you can iterate on it, and move from a complete system that works to a different complete system that works. This is much safer and faster, de-risking your progress.&lt;/li>&lt;li>You can put automated tests in place, to prevent any new functionality from breaking your desired system-wide properties.&lt;/li>&lt;/ol>&lt;p>Wait a minute though, what about &lt;a href="https://en.wikipedia.org/wiki/You_aren%27t_gonna_need_it">YAGNI&lt;/a>? Are we not wasting time trying to imagine a future that will never happen? Should we not face these challenges when and if they manifest? YAGNI has to do with the scope of your product, not the quality of your system.&lt;/p>&lt;p>What happens if a company doesn&amp;rsquo;t do this? Well, the problem is that the company will still need to design and build the aspects a walking skeleton covers. This is much harder once you have customers using your products, and while you&amp;rsquo;re building new features. You can still overdo a walking skeleton though, so it&amp;rsquo;s certainly a balance.&lt;/p>&lt;p>But you shouldn&amp;rsquo;t just ignore an aspect with the idea that you&amp;rsquo;ll think about it later. This is because the cost of doing something changes dramatically during the lifecycle of a system. Designing is about trade-offs, and these decisions can be hard to reverse.&lt;/p>&lt;p>So if you design your messaging infrastructure one way, ignore tenant isolation, and later on try to design for tenant isolation, you&amp;rsquo;ll find that the way your messaging infrastructure works will also have to change. And I&amp;rsquo;m talking about fundamental changes. These are only two aspects, but you&amp;rsquo;re actually looking at twenty or more, so when you change your messaging infrastructure, your monitoring, logging, and scaling will also have to change. When any of them changes, other aspects will have to change in turn, and so on. You can&amp;rsquo;t even do this in small steps, as the new parts don&amp;rsquo;t fit with the rest of the existing system.&lt;/p>&lt;p>How long does building a walking skeleton take? Well, there&amp;rsquo;s no standard answer to this. For tiny simple projects, it might only take a week or two. For large initiatives, it&amp;rsquo;ll typically take a competent cross-functional team of senior people four to six months to build a walking skeleton. Projects requiring multi-Cloud, multi-region, or dynamic provisioning will require a longer time for this. I&amp;rsquo;d say that each of these factors multiplies the time it takes to build a walking skeleton by a factor of 1.5.&lt;/p>&lt;p>Building a walking skeleton is not something you can parallelize much, because of the interdependencies involved in designing an end-to-end system. So, while raising the quality and expertise of the team involved will produce better results, increasing the number of people involved beyond ten or so will actually be counter-productive.&lt;/p>&lt;p>Mind that the goal of a walking skeleton is to dig out the unknown unknowns, and to design a system that can cope with them. It does not require that you build everything. You need to balance You Ain&amp;rsquo;t Going to Need It (YAGNI) with You Ain&amp;rsquo;t Going to Survive It (YAGSI). So you need to cover at least two very different languages for internationalization. But you shouldn&amp;rsquo;t go and build translation packs for every language known to humanity, just in case somebody might need it at some point.&lt;/p>&lt;h1 id="what-should-the-scope-be-for-a-walking-skeleton">What should the scope be for a walking skeleton?&lt;/h1>&lt;p>Let&amp;rsquo;s have a look at what a walking skeleton should cover. All these aspects should be addressed in a way that they all work together. So it doesn&amp;rsquo;t make sense to build one aspect and then move to the next one. Instead, build them in circles: plan them together, build a prototype, checks if things fit well, adjust, expand, etc.&lt;/p>&lt;p>I&amp;rsquo;ll try to group these aspects by category, but they&amp;rsquo;re all interdependent, so there&amp;rsquo;s no way to lay them out in dependency order. You should read through, understand how things could work together, and then decide whether you&amp;rsquo;d like the overall resulting system.&lt;/p>&lt;p>I&amp;rsquo;ll sometimes advise for specific approaches: take this with a pinch of salt, or feel free to ignore it, but changing one aspect will likely change how the overall system behaves, and what&amp;rsquo;s achievable. The important thing is that you consider all the aspects, make decisions, understand the trade-offs, and end up with a system that works well end-to-end.&lt;/p>&lt;h1 id="tenant-and-access-management">Tenant and access management&lt;/h1>&lt;h2 id="authentication">Authentication&lt;/h2>&lt;ul>&lt;li>Standard user journeys for onboarding, invitation, email verification, &lt;a href="https://en.wikipedia.org/wiki/Multi-factor_authentication">Multi-Factor Authentication&lt;/a> (MFA), login, and password reset.&lt;/li>&lt;li>Single Sign-On (SSO) options, including access via Google, etc. and external &lt;a href="https://en.wikipedia.org/wiki/Identity_provider">Identify Provider&lt;/a> (IDP) federation for B2B.&lt;/li>&lt;li>Choose between delegated or mapped identity for IDP federation cases. With delegated identity, the external IDP tells your IDP the identity of the user, and you take it at face value. With mapped identity, you map the identity from the external IDP in your own IDP. My advice here is to always map. In any case, you should ensure that when an identity is removed from an external IDP, it also gets removed from your IDP.&lt;/li>&lt;li>You&amp;rsquo;ll likely need to support both &lt;a href="https://openid.net/developers/how-connect-works/">OpenID Connect&lt;/a> (OIDC) and &lt;a href="https://en.wikipedia.org/wiki/SAML_2.0">Security Assertion Markup Language 2.0&lt;/a> (SAML 2.0), for IDP federation.&lt;/li>&lt;li>Tenants should be able to federate their IDP with yours programmatically, without you having to enable this manually for them.&lt;/li>&lt;li>Events for every time a user logs in for the first time, logs in, and logs out.&lt;/li>&lt;li>Whether to accept email aliases e.g., &lt;code>bruce.wayne+batman@gmail.com&lt;/code>.&lt;/li>&lt;li>A login challenge e.g., matching a password hash, something based on signing some bytes on both sides, etc.&lt;/li>&lt;li>Support for Personal Access Tokens for user accounts, and API keys for service accounts. Because orchestrated authentication workflows don&amp;rsquo;t play well with scripting.&lt;/li>&lt;li>Password storage implementation (Argon2 or similar to hash, salting, peppering, etc.).&lt;/li>&lt;li>You should allow users to view their active and historical sessions, including IP address, device type, location, and start and end dates. The user should also be able to invalidate all or specific sessions.&lt;/li>&lt;li>My strong advice is to stick to a battle-tested open-source solution, like &lt;a href="https://www.keycloak.org/">Keycloak&lt;/a>, or to go with a managed service like Okta. Out of the two, I&amp;rsquo;d go with hosting Keycloak, in general.&lt;/li>&lt;/ul>&lt;h2 id="token-management">Token management&lt;/h2>&lt;ul>&lt;li>Server-side sessions should be avoided.&lt;/li>&lt;li>Issuing, rotating, and verifying tokens, as part of your authentication workflows.&lt;/li>&lt;li>Certificate management is a prerequisite for this. You&amp;rsquo;ll need to support tenant-specific certificates, associated with tenant-specific subdomains e.g., &lt;code>&amp;lt;tenant&amp;gt;.&amp;lt;your-domain&amp;gt;.com&lt;/code>.&lt;/li>&lt;li>You should include a unique client-side session ID in your tokens, along with the start and end timestamps for the token&amp;rsquo;s validity.&lt;/li>&lt;li>The authorization roles and containing scopes should also be included in the token.&lt;/li>&lt;li>The duration of a token should be a maximum of 30m to 1 hour.&lt;/li>&lt;li>You also need to choose a digital signature scheme. &lt;a href="https://en.wikipedia.org/wiki/EdDSA">EdDSA&lt;/a>, ideally, rather than RSA or ECDSA, as it&amp;rsquo;s faster, requires shorter keys, and it&amp;rsquo;s considered more secure.&lt;/li>&lt;li>In terms of token structure, &lt;a href="https://en.wikipedia.org/wiki/JSON_Web_Token">JWT&lt;/a> is a safe bet.&lt;/li>&lt;/ul>&lt;h2 id="tenant-isolation">Tenant isolation&lt;/h2>&lt;ul>&lt;li>Tenants should be isolated from each other in terms of workflows.&lt;/li>&lt;li>The data for a tenant should be stored and encrypted separately from the data of other tenants.&lt;/li>&lt;li>Each tenant should be able to set up Single Sign-On with their own IDP, and to enable &lt;a href="https://www.cloudflare.com/learning/access-management/what-is-mutual-tls/">Mutual Transport Layer Security&lt;/a> (mTLS) and IP address range restrictions.&lt;/li>&lt;li>As mentioned above, you&amp;rsquo;ll need tenant-specific subdomains e.g., &lt;code>&amp;lt;tenant-name&amp;gt;.&amp;lt;your-root-domain&amp;gt;.com&lt;/code> for this. This needs to happen automatically when you onboard a new tenant.&lt;/li>&lt;li>If you offer environments, you&amp;rsquo;ll likely need infrastructure isolation to prevent the activity from one tenant to affect the workflows of another tenant.&lt;/li>&lt;li>Different tenants might require different encryption keys, physically segregated storage, and different locations for their data.&lt;/li>&lt;li>Data residency regulations require that data for customers in a Country is also going to be kept in that country (some Countries require &amp;ldquo;only&amp;rdquo; instead of &amp;ldquo;also&amp;rdquo;, but very few).&lt;/li>&lt;li>You&amp;rsquo;ll need to implement at least two tenants for each tenant type, as part of your walking skeleton.&lt;/li>&lt;/ul>&lt;h2 id="tenancy-model">Tenancy model&lt;/h2>&lt;ul>&lt;li>Your users will belong to different tenants, but a tenant will likely be too big to map to a single namespace.&lt;/li>&lt;li>You should allow your tenants to define more granular groups of users. You could choose to model these groups as projects, spaces, teams, or organizations.&lt;/li>&lt;li>My advice is to avoid choosing one model, and allow arbitrarily nestable containers, similar to folders. This way, a tenant can model their own organizational structures within your product, and allocate users and resources according to where they live in their own company.&lt;/li>&lt;li>The property of belonging to a container should cascade down the hierarchy. So if a tenant creates a &amp;ldquo;Global&amp;rdquo; folder, &amp;ldquo;US&amp;rdquo; and &amp;ldquo;EU&amp;rdquo; folders underneath it, and a &amp;ldquo;France&amp;rdquo; folder underneath &amp;ldquo;EU&amp;rdquo;, anything within &amp;ldquo;France&amp;rdquo; will also need to be within the &amp;ldquo;EU&amp;rdquo; and the &amp;ldquo;Global&amp;rdquo; folders.&lt;/li>&lt;li>These nestable containers can be used as containing scopes for authorization. So if a user has the &amp;ldquo;Project-Accessor&amp;rdquo; role within the &amp;ldquo;France&amp;rdquo; scope, they&amp;rsquo;ll see projects in France, but a different user with the same role in a different scope won&amp;rsquo;t.&lt;/li>&lt;li>These groupings are also useful for billing insights, so that billing contributions can be tagged with the containing scope they were produced in, allowing to roll them up hierarchically, according to the tenant&amp;rsquo;s own organizational structure.&lt;/li>&lt;li>As part of your walking skeleton, model at least two branches of the organizational hierarchy for each tenant, and do that for at least two separate tenants.&lt;/li>&lt;/ul>&lt;h2 id="authorization">Authorization&lt;/h2>&lt;ul>&lt;li>My advice is to model resources (the things your product allows to interact with), actions (what you can do with a resource), and container scopes (nestable groupings of resources).&lt;/li>&lt;li>I also advise defining policies as sets of action-resource permissions, and creating roles that are associated with bundles of policies. A user then has multiple roles, each associated with a containing scope. So when a user is in a given context, it has certain roles active in that context, and inherits some permissions through the policies associated with those roles, as a result.&lt;/li>&lt;li>Authorization should be checked and enforced once per invocation. If the grants for a user change, in-flight invocations should complete with the grants they had at the beginning of their processing.&lt;/li>&lt;li>You also need a way to query this, both on the front-end (to hide or gray out things the user cannot perform) and on the back-end (to enforce authorization rules).&lt;/li>&lt;li>On the front-end, you should retrieve the roles and the containing scopes from the user&amp;rsquo;s token, then look up the policies and permissions granted by those roles, from a dedicated endpoint.&lt;/li>&lt;li>You&amp;rsquo;ll need some standard available roles, plus the ability for your customers to define their own. To do this, you should describe what actions and policies each resource offers.&lt;/li>&lt;li>You should support mapping groups in your tenant&amp;rsquo;s Identity Provider to roles and scopes in your domain. So that if a customer&amp;rsquo;s employee is in the &amp;ldquo;UK&amp;rdquo; and the &amp;ldquo;developers&amp;rdquo; group, they might be associated with the &amp;ldquo;Standard User&amp;rdquo; role in the &amp;ldquo;UK&amp;rdquo; &amp;ldquo;Production&amp;rdquo; scope, and with the &amp;ldquo;Admin&amp;rdquo; role in the &amp;ldquo;UK&amp;rdquo; &amp;ldquo;Test&amp;rdquo; scope.&lt;/li>&lt;li>I recommend using a standard way of defining and querying authorization permissions and policies. &lt;a href="https://www.openpolicyagent.org/">Open Policy Agent&lt;/a> (OPA) is an excellent choice.&lt;/li>&lt;/ul>&lt;h1 id="high-level-architecture">High-level architecture&lt;/h1>&lt;h2 id="main-architectural-patterns">Main architectural patterns&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to choose whether to use orchestration (the logic is centralized in a controller, which dispatches calls to other services) or choreography (the logic is distributed, with each service reacting to what the other services are doing). &lt;a href="https://temporal.io/blog/to-choreograph-or-orchestrate-your-saga-that-is-the-question">A longer explanation of these two approaches&lt;/a>.&lt;/li>&lt;li>The choice between orchestration and choreography is for each workflow and subsystem. My advice is to adopt service choreography for every workflow and subsystem.&lt;/li>&lt;li>Choose whether to have as your source of truth the snapshot of the current state, or the history of the state changes. &lt;a href="https://www.eventstore.com/blog/what-is-event-sourcing">A longer explanation of these two approaches&lt;/a>. &lt;a href="https://developer.confluent.io/courses/event-sourcing/event-driven-vs-state-based/">Another explanation&lt;/a>.&lt;/li>&lt;li>This is also a local choice, so that each subsystem could choose a different approach. My advice is to record the history of state changes and use that as your source of truth, for all the subsystems.&lt;/li>&lt;li>I also recommend &lt;a href="https://en.wikipedia.org/wiki/Command_Query_Responsibility_Segregation">Command Query Responsibility Segregation&lt;/a> (CQRS). This means that commands (invocations that mutate the state without retrieving it) are processed by subsystems that are separate from the ones that process queries (invocations that retrieve the state without mutating it).&lt;/li>&lt;/ul>&lt;h2 id="event-driven-workflows">Event-driven workflows&lt;/h2>&lt;ul>&lt;li>Event-driven architectures work well with a partitioned distributed ledger as the backbone of event propagation.&lt;/li>&lt;li>&lt;a href="https://pulsar.apache.org/">Apache Pulsar&lt;/a> and &lt;a href="https://kafka.apache.org/">Apache Kafka&lt;/a> are both open-source and very popular. My recommendation is to go with Apache Pulsar, which I find superior to Kafka in every possible way. You should probably host your own Pulsar cluster, as it&amp;rsquo;s much easier than hosting Kafka, and much cheaper than a as-a-service commercial solution.&lt;/li>&lt;li>You should consider whether storing the events indefinitely within your distributed ledger (with tier storage backed by object storage, like S3 and Glacier), or whether using your ledger for propagation and a separate technology for events.&lt;/li>&lt;li>My advice is to have a store for events, like &lt;a href="https://www.eventstore.com/eventstoredb">EventStore&lt;/a> or even &lt;a href="https://www.postgresql.org/">Postgres&lt;/a>.&lt;/li>&lt;li>You&amp;rsquo;ll need a mechanism to handle &lt;a href="https://en.wikipedia.org/wiki/Personal_data">Personal Identifiable Information&lt;/a> (PII) with regard to GDPR&amp;rsquo;s Right To Be Forgotten. Storing PII as part of the events with field-level encryption using per end-user symmetric keys is a good approach. When the user needs to be forgotten, you can unlink the key, without losing data integrity.&lt;/li>&lt;li>Whenever you need to enforce domain rules that depend on the current state, you&amp;rsquo;ll have to choose between the &lt;a href="https://en.wikipedia.org/wiki/Inbox_and_outbox_pattern">outbox pattern&lt;/a> (write to a DB, then derive publishing an event), or &lt;a href="https://en.wikipedia.org/wiki/Compensating_transaction">the saga patter&lt;/a> (read from an eventually consistent view, decide in memory, publish an event, eventually update the view, rollback the changes and refund if there are conflicts). My advice here is to use sagas, as the outbox pattern is tricky and constitutes a severe limitation in throughput.&lt;/li>&lt;li>As your system is fully asynchronous, message-based, and reactive, you can use the &lt;a href="https://medium.com/@mahernaija/messaging-patterns-cf4bc5b164cf">async request-reply pattern&lt;/a> to respond to queries synchronously.&lt;/li>&lt;li>My advice is to use &lt;a href="https://en.wikipedia.org/wiki/NATS_Messaging">NATS&lt;/a> for this. You generate a unique response NATS topic, subscribe to it, emit an event tagged with the response topic, await for a downstream service to publish the outcome of your workflow to NATS, and finally respond to the open socket. MQTT works well for this too, but it&amp;rsquo;s quite a heavy technology to deploy.&lt;/li>&lt;/ul>&lt;h2 id="api-style">API style&lt;/h2>&lt;ul>&lt;li>In terms of API style, you&amp;rsquo;ll have to pick one and stick to it for your whole API surface. Examples of API styles include &lt;a href="https://en.wikipedia.org/wiki/REST">REST&lt;/a>, &lt;a href="https://en.wikipedia.org/wiki/GraphQL">GraphQL&lt;/a>, and &lt;a href="https://en.wikipedia.org/wiki/Remote_procedure_call">RPC&lt;/a>.&lt;/li>&lt;li>My advice is to adopt an RPC style. GraphQL and REST both operate at data manipulation level, but what you want is domain-specific actions. I&amp;rsquo;m not talking about &lt;a href="https://grpc.io/">gRPC&lt;/a> by the way, as it brings its own protocol tweaks and doesn&amp;rsquo;t work well with many existing tools. Go over standard HTTP 2, and use either &lt;a href="https://en.wikipedia.org/wiki/JSON">JSON&lt;/a> or &lt;a href="https://en.wikipedia.org/wiki/Avro">Avro&lt;/a> as data format. I&amp;rsquo;d recommend Avro, but JavaScript in the browser supports it poorly, so even JSON works fine.&lt;/li>&lt;li>So use the POST HTTP method, with the type of the invocation as part of the path, and choose whether to differentiate commands and queries. Examples include &lt;code>POST /commands/tranfer-account-ownership&lt;/code>, &lt;code>POST /queries/retrieve-available-balance&lt;/code>, and &lt;code>POST /invocations/tranfer-account-ownership&lt;/code>.&lt;/li>&lt;li>You can specify caching instructions for POST responses, but my advice is to avoid caching responses as much as possible. If you really need that performance, subscribe to changes on the client side.&lt;/li>&lt;li>This approach enables bulk invocations endpoints e.g. &lt;code>POST /commands&lt;/code>, &lt;code>POST /queries&lt;/code> and &lt;code>POST /invocations&lt;/code>, so a client can send multiple invocations in one exchange with the server.&lt;/li>&lt;li>This way, it&amp;rsquo;s easy to record all the invocations that happened, whether commands or queries, to provide full auditability.&lt;/li>&lt;/ul>&lt;h2 id="server-push-data">Server-push data&lt;/h2>&lt;ul>&lt;li>Your back-end will need to sometimes push data to your client applications. This means the server will initiate a communication with the client, rather than the other way around.&lt;/li>&lt;li>You should either host a solution, like an &lt;a href="https://mqtt.org/">MQTT&lt;/a> cluster, like &lt;a href="https://www.emqx.io/">EMQX&lt;/a>, or going for a commercial solution, like &lt;a href="https://ably.com/">Ably&lt;/a>.&lt;/li>&lt;li>My advice is to avoid websockets, as they introduce connection stickiness (you need to know which socket to use to send a message to a user), and they&amp;rsquo;re too low level.&lt;/li>&lt;li>A user shouldn&amp;rsquo;t be able to subscribe to information they shouldn&amp;rsquo;t see (data belonging to other tenants or users), so you&amp;rsquo;ll need authentication and authorization.&lt;/li>&lt;li>You&amp;rsquo;ll need to structure your topics carefully, balancing authorization needs and the effort involved in pushing an update. So a topic per user makes authorization trivial, but pushing updates a nightmare. A single topic is the opposite. You&amp;rsquo;ll need a balance.&lt;/li>&lt;li>My advice is to use a hosted open-source solution when you plan on sending server-pushed information often, as the cost builds up with commercial solutions.&lt;/li>&lt;/ul>&lt;h2 id="gateway-and-services">Gateway and services&lt;/h2>&lt;ul>&lt;li>Choose whether to leverage an API gateway, or to go without one. My advice would be to use one, so you can centralize token signature verification, authorization enforcement, parsing of tracing information, &lt;a href="https://en.wikipedia.org/wiki/Cross-origin_resource_sharing">CORS&lt;/a> handling, &lt;a href="https://en.wikipedia.org/wiki/TLS_termination_proxy">TLS termination&lt;/a>, and other similar aspects. I also recommend hosting an open-source or a commercial solution, rather than buying a service.&lt;/li>&lt;li>You&amp;rsquo;ll want to create templates for the various types of services that appear in your topology. Command endpoints (receive and emit commands), query endpoints (serve queries), event processors (receive and emit events), event sinks (receive and consume events, typically integrating with third-party technologies, or updating materialized views).&lt;/li>&lt;/ul>&lt;h1 id="correlation">Correlation&lt;/h1>&lt;h2 id="invocation-context">Invocation context&lt;/h2>&lt;ul>&lt;li>Every invocation should be contextualised with metadata about the invocation itself.&lt;/li>&lt;li>This context should be part of your software models, not something implicit and accessed outside the normal call stack. You should be able to base your behavior on this information, to log it, and to include it in the events you publish.&lt;/li>&lt;li>The context should contain information about the access, the trace, and the toggles associated with the invocation.&lt;/li>&lt;li>The toggles associated to an invocation are a set of key-value pairs, where the key is a feature toggle ID, and the value is the type-safe value for that feature. This can be used for A/B testing, to bump the logging level for an invocation, and for many other use cases. Mind that these toggles are per-invocation, so they shouldn&amp;rsquo;t contain things like product tiers, etc.&lt;/li>&lt;li>The access should include authorization information, provenance information (client IP address, parsed user agent, etc.). It should also indicate whether the invocation was unauthenticated (for operations exposed outside of login) or authenticated. For authenticated accesses, information about the actor should also be included. This should indicate the account ID, whether it&amp;rsquo;s a user or a service account, the tenant the actor belongs to, the locale of the user, and the authentication mechanism (including client-side session ID or the hash of the API key used). The actor model should also support impersonation and acting on behalf of other accounts.&lt;/li>&lt;li>The trace should allow correlating invocations that are logically related. A good way of achieving this is to generate some initial unique IDs (e.g., &lt;a href="https://github.com/ulid/spec">ULID&lt;/a>s) on the client side: 1 for the action itself e.g., a button press, plus 1 for each invocation to the back-end this action results in (typically one). After that, the gateway should generate an originating trace ID, and these 3 pieces of information should be immutable throughout the chain of invocations. Each service that processes the invocation should receive a parent invocation ID, generate an invocation ID, and send this last ID as parent invocation ID for downstream calls.&lt;/li>&lt;li>This way humans and systems can easily identify retries and invocation that are logical duplicates of previous invocations.&lt;/li>&lt;li>The information in the invocation context allows determining an idempotency key: this one is typically the external invocation ID, within a namespace composed of the tenant ID and the actor ID (so that different actors and tenants cannot interfere with each other in terms of idempotency).&lt;/li>&lt;li>An application should receive the full invocation context from the gateway (typically encoded as JSON and to Base64 in a header), fork it (generate a unique invocation ID, and move the received invocation ID as new parent invocation ID), add it to the log stack, pass it as an argument (or as a context parameter) to all functions, include this information in every event, and send this as a header as part of any downstream HTTP request.&lt;/li>&lt;li>Every event and message should also have their own context information, which is the ID of the originating and the parent events and messages. So each event has an event context, which includes the invocation context, and event correlation information. Each message (the event might be in the payload), will also contain message correlation information.&lt;/li>&lt;/ul>&lt;h2 id="idempotency">Idempotency&lt;/h2>&lt;ul>&lt;li>The invocation context gives you an idempotency ID.&lt;/li>&lt;li>You can use this ID to ensure that logically duplicated calls do not cause duplicated side effects, including state changes and calls to external systems.&lt;/li>&lt;li>For SQL databases, you can use a transaction that inserts this ID into a dedicated table of previously seen invocations, with a unique constraint. As part of this transaction, you can do whatever that invocation needs to do from a business perspective. If the transaction fails because of the idempotency constraint, you should consider the operation successful, and proceed with anything else you would normally do if it completed regularly.&lt;/li>&lt;li>For external APIs, some will allow you to specify an idempotency key. In this case, you simply pass your external invocation ID, and the 3rd-party system will take care of idempotency on its end. This works the same for your system, since your clients can choose to repeat the external invocation ID, if they want to retry an invocation in a safe way.&lt;/li>&lt;li>Some 3rd-party APIs e.g., email senders, won&amp;rsquo;t allow you to specify an idempotency ID. In this case, you can use a SQL database to filter out duplicates, as per above except for skipping the call to the external system in case of a uniqueness constraint failure with the idempotency ID. This approach does not guarantee idempotency, because if you cannot transactionally invoke the 3rd-party API and your database, so if you crash in between, you can end up with duplicate invocations.&lt;/li>&lt;/ul>&lt;h1 id="user-facing-applications">User-facing applications&lt;/h1>&lt;h2 id="internationalization">Internationalization&lt;/h2>&lt;ul>&lt;li>All strings should be represented using the UTF-16 or UTF-8 charsets. UTF-8 can represent any character, but it&amp;rsquo;s optimized for text that is mostly ASCII. So it depends on the intensity of this requirement. In any case, stick to UTF-8 at a minimum, even if you don&amp;rsquo;t currently have any internationalization requirements.&lt;/li>&lt;li>The frontend applications should externalize all text content, so that a different value can be used based on the browser&amp;rsquo;s locale. When the locale changes in the browser, this should also be changed.&lt;/li>&lt;li>The backend should return these internationalization keys, rather than text in English, for everything.&lt;/li>&lt;li>The actual values should be packaged in the app itself, falling back to a &lt;a href="https://en.wikipedia.org/wiki/Content_delivery_network">Content Delivery Network&lt;/a> (CDN) lookup. This allows trading off chattiness and app size, and changing text without having to release.&lt;/li>&lt;li>All communications e.g., emails and in-app notifications should respect the locale of the user, falling back to a default locale, if the user&amp;rsquo;s locale is not currently supported.&lt;/li>&lt;li>If you plan to support Arab countries, or Asian countries like China and Japan, you should also change the page layout based on the user locale, as some of these languages are laid out differently.&lt;/li>&lt;/ul>&lt;h2 id="in-app-notifications-of-changes">In-app notifications of changes&lt;/h2>&lt;ul>&lt;li>You should model new features, marketing communications, etc. as code, and notify your users about them both in-app and through emails. The notifications should link to a browsable changelog of all historical changes to your products.&lt;/li>&lt;li>This changelog should be an append-only log of changes, so that a user can browse and search them. Each item in this list should take the user to a page with more details about that specific change.&lt;/li>&lt;/ul>&lt;h2 id="in-app-audit-log">In-app audit log&lt;/h2>&lt;ul>&lt;li>Users should be able to browse, filter, and inspect a changelog with all the actions they ever performed.&lt;/li>&lt;li>If a user has the specific permission, they should be able to also see in the changelog the actions performed by other users.&lt;/li>&lt;li>Each item in the audit log should contain the full context of the invocation represented by that item, with dates, access, etc.&lt;/li>&lt;/ul>&lt;h2 id="displaying-and-reporting-errors">Displaying and reporting errors&lt;/h2>&lt;ul>&lt;li>You should display meaningful error messages when something goes wrong, in the user&amp;rsquo;s locale (with fallback, if not supported). These shouldn&amp;rsquo;t be protocol-level error messages e.g., 403 Forbidden, but properly crafted messages, explaining what went wrong, and what the user should do about it.&lt;/li>&lt;li>Some types of errors should offer the user the option to retry, automatically using the same external invocation ID for idempotency.&lt;/li>&lt;li>All unexpected errors should ask the user to report the error, automatically adding correlation information about the invocation. Reporting an error should create an entry within an in-app error investigations section, and retry the invocation with bumped up logging level (using a toggle) and same external invocation ID (for idempotency).&lt;/li>&lt;li>This should alert your back-office. Once the error is fixed, a back-office user can provide information about what happened and what was done about it. The open investigation should be updated, and the user should receive an in-app and an email notification.&lt;/li>&lt;/ul>&lt;h2 id="accessibility">Accessibility&lt;/h2>&lt;ul>&lt;li>If you have any accessibility requirements, and probably even if you don&amp;rsquo;t, you should ensure your client applications are accessible.&lt;/li>&lt;li>You should create automated tests to prevent regressions (using tools like &lt;a href="https://wave.webaim.org/">Wave&lt;/a>, &lt;a href="https://www.deque.com/axe/">Axe&lt;/a>, and &lt;a href="https://pa11y.org/">Pa11y&lt;/a>, &lt;a href="https://opensource.com/article/23/2/automated-accessibility-testing)">https://opensource.com/article/23/2/automated-accessibility-testing)&lt;/a>, so that what&amp;rsquo;s accessible doesn&amp;rsquo;t lose this property at some point in the future.&lt;/li>&lt;/ul>&lt;h2 id="in-app-and-off-app-learning">In-app and off-app learning&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to walk your users through your product.&lt;/li>&lt;li>This can be a mix of in-app resources e.g., a guided walkthrough and your product&amp;rsquo;s documentation, and off-app resources e.g., videos on YouTube.&lt;/li>&lt;li>In any case, you should list these resources in-app, and guide the users through their journey.&lt;/li>&lt;li>My advice is to use an existing product tour tool, rather than trying to roll your own. &lt;a href="https://www.storylane.io/blog/11-best-product-tour-software-for-saas">Here&amp;rsquo;s a selection of some options for this&lt;/a>.&lt;/li>&lt;/ul>&lt;h2 id="history-of-changes-for-each-entity">History of changes for each entity&lt;/h2>&lt;ul>&lt;li>Your product will expose specific entities, allowing users to manipulate these entities to achieve their goals.&lt;/li>&lt;li>You should ensure that your users can browse and search the history of all changes that happened on any one of such entities.&lt;/li>&lt;li>An example might be that, if your product allows users to create projects, any user with permissions should be able to see the whole history of changes on a specific project: who created it and when, the original data, the log of changes, each including who did what, when, and how.&lt;/li>&lt;/ul>&lt;h2 id="multiplatform-strategy">Multiplatform strategy&lt;/h2>&lt;ul>&lt;li>If you ever need more than platform e.g., iOS, Android, and web, you&amp;rsquo;ll need to decide between leveraging a multiplatform approach (like &lt;a href="https://react.dev/">React&lt;/a>/&lt;a href="https://reactnative.dev/">React Native&lt;/a>, &lt;a href="https://kotlinlang.org/docs/multiplatform.html">Kotlin Multiplatform&lt;/a>, etc.), and adopting separate codebases and technologies for each channel.&lt;/li>&lt;li>In B2B, it almost always makes sense to go with a multiplatform approach. In B2C, your users will tend to be mobile-first, so going native might sometimes be preferable.&lt;/li>&lt;li>However, this is not a rule, so make an explicit decision, keeping in mind the number of specialists you&amp;rsquo;ll need, and how exacting are your platform-specific requirements.&lt;/li>&lt;/ul>&lt;h2 id="ui-components">UI components&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll want to curate a set of UI components to avoid duplication and achieve a consistent user experience, throughout your applications.&lt;/li>&lt;li>You should be able to compose these into higher-order components and workflows.&lt;/li>&lt;li>You should be able to test each component in isolation.&lt;/li>&lt;li>These components should be versioned and maintained as a library.&lt;/li>&lt;/ul>&lt;h2 id="modules-features-and-billing">Modules, features, and billing&lt;/h2>&lt;ul>&lt;li>Your products will contain various modules and features.&lt;/li>&lt;li>You&amp;rsquo;ll need to model these explicitly in your code, so that a new module can be released without having to release the front-end applications.&lt;/li>&lt;li>You should be able to bill based on a combination of a base tier, plus a consumption-based contribution. In any case, you should be able to clearly explain your bills, with hierarchical roll-ups based on nested containing scopes.&lt;/li>&lt;li>Whenever a tenant would have exceeded their maximum quota of operations, you should be able to automatically disable further actions from that tenant, until they expand their subscription or a new billing cycle starts.&lt;/li>&lt;/ul>&lt;h2 id="instrumentation-and-attribution">Instrumentation and attribution&lt;/h2>&lt;ul>&lt;li>You should instrument your products, so that you can learn what your users are clicking, doing, etc.&lt;/li>&lt;li>These events should be captured within the client applications, and sent to the server or to a 3rd-party service.&lt;/li>&lt;li>Instrumentation should include both low-level actions e.g., clicks, typed text, etc. and high-level intentions e.g., logged in, created a project, changed the permissions of another user, viewed the list of projects, etc.&lt;/li>&lt;li>You should also be able to attribute events to user-specific toggles, to gauge the impact your tweaks had on the KPIs they hoped to improve.&lt;/li>&lt;li>My advice is to use an external tool for both instrumentation and attribution. Examples could be &lt;a href="https://amplitude.com/">Amplitude&lt;/a> for both, or &lt;a href="https://www.pendo.io/">Pendo&lt;/a> for instrumentation and &lt;a href="https://www.split.io/">Split&lt;/a> for attribution.&lt;/li>&lt;/ul>&lt;h2 id="requirements-and-entitlements">Requirements and entitlements&lt;/h2>&lt;ul>&lt;li>Some features, modules, or products will come with additional requirements.&lt;/li>&lt;li>Even just accessing your platform will likely have some basic requirements e.g., a valid email address.&lt;/li>&lt;li>You should model what requirements each feature, module, product, or platform needs, before a user can access it, within your code.&lt;/li>&lt;li>Each requirement could be satisfied by various alternative entitlements. You should model which entitlements satisfy a requirement. This can vary based on the country, etc.&lt;/li>&lt;li>An example might be that a valid email address requirement could be satisfied by a passed email verification check, or by receiving a corporate email address from a federated identity provider.&lt;/li>&lt;li>Users should be associated with the set of entitlements they already have. A product with requirements that already satisfied by the entitlements a user has shouldn&amp;rsquo;t ask the user to do anything else.&lt;/li>&lt;li>You should be capable of disabling functionality the user hasn&amp;rsquo;t qualified to use yet, of explaining which requirements aren&amp;rsquo;t met yet, of showing the available ways of qualifying for each requirement, of running the verification step for the entitlement chosen by the user, and finally of unlocking the feature when all the requirements are met.&lt;/li>&lt;li>Requirements and entitlements should be dynamic, sent by the server to the client as data. Your client applications shouldn&amp;rsquo;t know which requirements a product or feature needs. The same product might have different requirements across different countries. And a requirement might be met by different entitlements, in different countries.&lt;/li>&lt;li>This works very well with components. So a requirement might be &amp;ldquo;having accepted the terms and conditions with ID 123&amp;rdquo;, an entitlement might be &amp;ldquo;digitally signed a document showing the terms and conditions with ID 123&amp;rdquo;, and you could have a component for this entitlement, so that any time a product or feature needs terms and conditions signed, you can re-use the component as part of that workflow.&lt;/li>&lt;/ul>&lt;h2 id="documents-and-reports">Documents and reports&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll likely need to develop the ability to produce PDFs for terms and conditions, and for reports.&lt;/li>&lt;li>Legal documents should be stored in an immutable way, rather than being produced again in the future.&lt;/li>&lt;li>So if you&amp;rsquo;re a bank and a user produces a statement of their transactions, you should store that statement, so that the user is guaranteed to access the same statement in the future, even if you change your business logic.&lt;/li>&lt;/ul>&lt;h2 id="sdks">SDKs&lt;/h2>&lt;ul>&lt;li>You should abstract all communications to and from the back-end with client-side SDKs.&lt;/li>&lt;li>Each SDK should be built, tested, versioned, and released as a stand-alone library.&lt;/li>&lt;li>Each SDK should have a contract-only module, and various implementations e.g., HTTP-based + MQTT for server-pushed communications, or an in-memory implementation for local testing.&lt;/li>&lt;li>You should test your client applications against an in-memory version of the SDK, not against an environment hosting your back-end.&lt;/li>&lt;li>You should use reactive-streams e.g., &lt;a href="https://rxjs.dev/">RxJS&lt;/a> or &lt;a href="https://mobx.js.org/README.html">MobX&lt;/a> to allow the client to subscribe to server-pushed data.&lt;/li>&lt;li>You should use the SDK to write tests that don&amp;rsquo;t test the UI e.g., smoke tests you can run periodically in production.&lt;/li>&lt;li>You should have test extensions for your SDKs, allowing you to easily create preconditions that wouldn&amp;rsquo;t make sense as normal operations.&lt;/li>&lt;li>You might want to give your client-facing SDKs to your customers. You shouldn&amp;rsquo;t give your customers the test extensions.&lt;/li>&lt;/ul>&lt;h1 id="infrastructure">Infrastructure&lt;/h1>&lt;h2 id="only-one-long-lived-environment">Only one long-lived environment&lt;/h2>&lt;ul>&lt;li>You should have only one long-lived environment: production. So no test, UAT, or staging environments.&lt;/li>&lt;li>If you support dynamic environments, every such environment should be production for you. If your customers want a test environment, this should still be a production environment for you. To understand this better, think about how Cloud providers work.&lt;/li>&lt;li>Learn how to test locally, and leverage internal tenants to test and demo your changes before releasing them to external tenants.&lt;/li>&lt;li>Use ephemeral short-lived environments to perform intense tests e.g., stress performance tests, or to test large-scale infrastructure changes.&lt;/li>&lt;li>These ephemeral environments should be exact copies of the production environment, with masked sensitive tenant data, and you should be able to spin them up and down programmatically.&lt;/li>&lt;/ul>&lt;h2 id="infrastructure-as-code">Infrastructure as code&lt;/h2>&lt;ul>&lt;li>You should describe your entire infrastructure and tenant configuration in a declarative way, as code.&lt;/li>&lt;li>Changes in the code should trigger changes in your infrastructure.&lt;/li>&lt;li>You can use various approaches for this, including &lt;a href="https://www.pulumi.com/">Pulumi&lt;/a> and &lt;a href="https://opentofu.org/">Terraform/OpenTofu&lt;/a>.&lt;/li>&lt;li>My advice is to use Pulumi, as it allows you to dynamically declare resources based on runtime data. You can also create your own types and functions, and it&amp;rsquo;s easier to compose.&lt;/li>&lt;li>You should monitor your infrastructure, checking for divergences between the declared and the actual state. A configuration drift management tool can help you spot these inconsistency, raise alerts, and even roll-back any undeclared changes.&lt;/li>&lt;li>Structure a mechanism for requesting and granting temporary and audited access (read or write) to resources. These requests need to be approved and recorded. Nobody should have direct access to any database, network, console, or anything.&lt;/li>&lt;li>In terms of provisioning, a fully scripted approach works well, except when infrastructure needs to be provisioned as part of some users&amp;rsquo; actions. In this case, you&amp;rsquo;ll need dynamic provisioning, and Terraform and Pulumi can leave you with an inconsistent state in case of failure. &lt;a href="https://kubernetes.io/docs/concepts/extend-kubernetes/operator/">Kubernetes Operators&lt;/a> are a better approach in this case.&lt;/li>&lt;li>My advice would be to use Pulumi to set up &lt;a href="https://kubernetes.io/">Kubernetes&lt;/a> and install a bootstrap Operator, and then use &lt;a href="https://kubernetes.io/docs/concepts/extend-kubernetes/operator/">Kubernetes Operators&lt;/a> for everything else, even when you don&amp;rsquo;t have dynamic infrastructure provisioning requirements.&lt;/li>&lt;/ul>&lt;h2 id="configuration-management">Configuration management&lt;/h2>&lt;ul>&lt;li>Your applications will need configuration variables, secret keys, certificates, passwords, etc.&lt;/li>&lt;li>These might depend on the specific tenant the invocation belongs to. Two tenants might use separate databases, each with their own credentials.&lt;/li>&lt;li>Whenever you can, avoid injecting these variables into your applications at all. Instead, use side-cars that proxy the connections or do the work that requires the secrets.&lt;/li>&lt;li>When not feasible, use a standard way of injecting configuration, like &lt;a href="https://kubernetes.io/docs/concepts/configuration/configmap/">Kubernetes ConfigMaps&lt;/a>, or file-based configuration exposed to the apps through ephemeral volumes.&lt;/li>&lt;li>In terms of storing secrets, you&amp;rsquo;ll want something audited, segregated, and secure, like &lt;a href="https://www.hashicorp.com/products/vault">Hashicorp Vault&lt;/a>. You don&amp;rsquo;t want each application to interact with Vault directly, though. So either expose the secrets in Vault through &lt;a href="https://kubernetes.io/docs/concepts/configuration/secret/">Kubernetes Secrets&lt;/a>, or pass them to your applications inside files on ephemeral volumes.&lt;/li>&lt;li>You should ensure that whenever a configuration value changes, your applications can pick up the change. I recommend rebooting the containers in this case, as it&amp;rsquo;s usually not a big deal.&lt;/li>&lt;li>All credentials should be service-specific, never shared by multiple applications.&lt;/li>&lt;/ul>&lt;h2 id="certificate-management">Certificate management&lt;/h2>&lt;ul>&lt;li>Your products will use certificates, and those certificates will need rotation, etc.&lt;/li>&lt;li>In multi-tenant contexts, each tenant will likely end up with its own subdomain and its own certificates.&lt;/li>&lt;li>Figure out how to manage dynamic certificates: either by issuing them through a root certificate, or by rolling your own &lt;a href="https://en.wikipedia.org/wiki/Certificate_authority">Certificate Authority&lt;/a>.&lt;/li>&lt;li>Whenever a tenant or an infrastructure zone requires its own certificate, you should be able to issue it programmatically.&lt;/li>&lt;/ul>&lt;h2 id="internet-edge-solution">Internet Edge Solution&lt;/h2>&lt;ul>&lt;li>An internet edge solution, like &lt;a href="https://www.fastly.com/">Fastly&lt;/a> or &lt;a href="https://www.cloudflare.com/">CloudFlare&lt;/a> is almost always a great idea.&lt;/li>&lt;li>You get protection from DDoS attacks, and against other malicious attempts.&lt;/li>&lt;li>Also, you can allow your tenants to enable mTLS and IP address range restrictions programmatically, by calling the internet edge solution&amp;rsquo;s API as part of handling such a command on your back-end.&lt;/li>&lt;li>My advice is to use &lt;a href="https://www.fastly.com/">Fastly&lt;/a>, as it&amp;rsquo;s usually powerful and easy to integrate with.&lt;/li>&lt;/ul>&lt;h2 id="cloud-providers">Cloud providers&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to decide whether having single Cloud provider availability and business continuity is enough.&lt;/li>&lt;li>If so, then my advice is to stick with &lt;a href="https://cloud.google.com/">Google Cloud Platform&lt;/a> (GCP). It&amp;rsquo;s cheaper than the others, and offers great essential products e.g., the managed Kubernetes service.&lt;/li>&lt;li>If you need a multi-Cloud approach, you&amp;rsquo;ll need to figure out whether active-passive-cold is enough, or whether you need active-passive-hot or active-active.&lt;/li>&lt;li>In any case, the way this should work is that, when your active Cloud provider has an issue, you should be able to switch to another Cloud provider with 1 click.&lt;/li>&lt;li>You should aim at being able to switch without losing any data, and with zero downtime (that&amp;rsquo;s where active-passive-hot and active-active come into play).&lt;/li>&lt;li>Active-passive-hot and active-active multi-Cloud workflows are rare and hard, but they can be achieved in a similar way active-active multi-region workflows work.&lt;/li>&lt;li>You could deploy a messaging cluster on each Cloud provider, and then replicate each command across all clusters, ensuring that each replicated command ends up having the same offset across different clusters. Apache Pulsar can help with this, with its geo-replication features. Once you have this, you can enable/disable the side effects in your passive Cloud provider, but it&amp;rsquo;ll be ready and eventually consistent with the active one.&lt;/li>&lt;li>If you need a multi-Cloud approach, you should use at least three Cloud providers as part of your walking skeleton. This is because you&amp;rsquo;ll need quorum to recover from a temporary failure, so you&amp;rsquo;ll need three Cloud providers at a minimum, if one is not enough.&lt;/li>&lt;/ul>&lt;h2 id="regions-and-availability-zones">Regions and Availability Zones&lt;/h2>&lt;ul>&lt;li>Depending on your availability and disaster recovery requirements, you&amp;rsquo;ll need your services across multiple Availability Zones, or even across multiple Regions. &lt;a href="https://cloud.google.com/compute/docs/regions-zones">More information about Regions and Zones&lt;/a>.&lt;/li>&lt;li>Active-active and active-passive-hot multi-region workflows are hard, just like it&amp;rsquo;s hard to get these working well across multiple Cloud providers.&lt;/li>&lt;li>The difference is that multiple Cloud providers introduce higher management complexity, while multiple Regions introduce higher latency involved in your replication.&lt;/li>&lt;li>If you have multi-region requirements, you should test in at least 3 regions as part of your walking skeleton. This is because you usually need quorum to recover from a temporary failure, so the minimum regions you&amp;rsquo;ll need is 3, when 1 is not enough.&lt;/li>&lt;/ul>&lt;h2 id="service-registry">Service registry&lt;/h2>&lt;ul>&lt;li>You should have a service registry, listing all your existing services.&lt;/li>&lt;li>Each service should have an identity in the form of a unique identifier e.g., a &lt;a href="https://github.com/ulid/spec">ULID&lt;/a>.&lt;/li>&lt;li>Each service should show which infrastructure resources they depend on, along with the libraries and versions they use.&lt;/li>&lt;li>You should be able to see all services that use a dependency version affected by a vulnerability.&lt;/li>&lt;li>Every service should also link to its codebase on version control.&lt;/li>&lt;/ul>&lt;h2 id="messaging">Messaging&lt;/h2>&lt;ul>&lt;li>Your messaging infrastructure will need authorization, auto-scaling, replication, partitioning, etc.&lt;/li>&lt;li>You should have a service registry in place, and a way of authenticating services. mTLS is a good way of doing this, if you issue per-service certificates, with rotation, using sidecars to proxy the communications.&lt;/li>&lt;li>In terms of authorization, &lt;a href="https://docs.streamnative.io/docs/access-control">ACLs&lt;/a> work well, but a custom &lt;a href="https://pulsar.apache.org/docs/next/security-extending/">Authorization Provider&lt;/a> using &lt;a href="https://www.openpolicyagent.org/">Open Policy Agent&lt;/a> (OPA) is better. Both Pulsar and Kafka support this. The way it works is that you can declare a custom Authorization Provider as a plugin extension, and your broker will ask it whether a service is allowed to publish or subscribe to a topic, caching the result afterward.&lt;/li>&lt;li>Partitioning and &lt;a href="https://docs.datastax.com/en/streaming/kaap-operator/0.1.0/">cluster auto-scaling&lt;/a> are much easier with Pulsar than they are with Kafka, because Apache Pulsar&amp;rsquo;s brokers are stateless. You&amp;rsquo;ll need to template both aspects as part of your walking skeleton.&lt;/li>&lt;li>You should also have a company-wide registry of topics and schemata (typically in &lt;a href="https://avro.apache.org/">Apache Avro&lt;/a>). So that each developer should know what topics exist, what are they for, and what schema is published to that topic.&lt;/li>&lt;li>Your cluster should be configured to reject messages whose payload is backward or forward incompatible with the schema set for the topic. &lt;a href="https://pulsar.apache.org/docs/3.1.x/schema-understand#schema-evolution">An overview of how this works in Pulsar&lt;/a>.&lt;/li>&lt;/ul>&lt;h2 id="service-to-service-communications">Service-to-service communications&lt;/h2>&lt;ul>&lt;li>Ideally no two services should ever know of each other&amp;rsquo;s existence, exclusively publishing and receiving events instead.&lt;/li>&lt;li>If you must have peer-to-peer communications, service mesh will help manage this.&lt;/li>&lt;li>By using peer-to-peer communications you lose a lot, in terms of backpressure, security, extensibility, etc. so think really carefully whether you really have to.&lt;/li>&lt;li>For external services, you&amp;rsquo;ll need peer-to-peer communications anyway. My advice is to structure proxy services for 3rd-party dependencies, so that your applications are not concerned with certificate management, credentials, etc.&lt;/li>&lt;/ul>&lt;h2 id="outbound-invocations">Outbound invocations&lt;/h2>&lt;ul>&lt;li>If your applications initiate outbound invocations to a tenant&amp;rsquo;s infrastructure, you&amp;rsquo;ll need to somehow prove these are genuine and originated from the right environment.&lt;/li>&lt;li>For webhooks, you should consider &lt;a href="https://httpwg.org/http-extensions/draft-ietf-httpbis-message-signatures.html">HTTP Message Signatures&lt;/a>.&lt;/li>&lt;li>Each tenant-specific area of your infrastructure should generate a key pair (with rotation) used to sign outbound requests.&lt;/li>&lt;li>The receiver should retrieve the public key from a well-known endpoint exposed under the tenant-specific subdomain (so protected by mTLS, IP address range restrictions, etc.), and use the key to verify the signature.&lt;/li>&lt;li>Like for calls to external services, my advice is to use proxy services to produce and attach HTTP Message Signatures to outbound invocations, so that your applications are not concerned with certificate management, credentials, etc.&lt;/li>&lt;/ul>&lt;h2 id="autoscaling-for-services">Autoscaling for services&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll want CPU-based auto-scaling groups, for command and query endpoints, as their driven by push-based unpredictable external traffic.&lt;/li>&lt;li>For event processors and event sinks, CPU-based autoscaling makes no sense, as these process messages out of a queue. Instead, you&amp;rsquo;ll want to scale their consumer groups based on the derivative of their queue size: if they&amp;rsquo;re falling behind the producers, ramp up, and if they&amp;rsquo;re catching up fast, ramp down.&lt;/li>&lt;/ul>&lt;h2 id="monitoring">Monitoring&lt;/h2>&lt;ul>&lt;li>You should have dashboards and alerts for both product and infrastructure-related aspects.&lt;/li>&lt;li>In terms of infrastructure, things like CPU, memory, number of inbound and outbound connections, error rate, available disk space, etc.&lt;/li>&lt;li>About product aspects, stuck workflows that are not progressing through the steps after a given time (using sagas for this), brute-force login attempts, suspicious patterns, etc.&lt;/li>&lt;li>You should also have health checks and readiness checks for each service, along with an external availability check service e.g., &lt;a href="https://www.pingdom.com/">Pingdom&lt;/a>, to check whether your platform is available to the outside world.&lt;/li>&lt;li>Any error-level log entry should trigger an alert to be investigated.&lt;/li>&lt;li>My advice is to use open-source tools for monitoring and for dashboards e.g., &lt;a href="https://prometheus.io/">Prometheus&lt;/a> and &lt;a href="https://grafana.com/">Grafana&lt;/a>, as opposed to commercial solutions like &lt;a href="https://www.datadoghq.com/">Datadog&lt;/a>, as these can get very expensive fast.&lt;/li>&lt;/ul>&lt;h2 id="releases-and-deployments">Releases and deployments&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need automatic canary deployments, gradually incrementing the amount of traffic directed to the upcoming version, based on comparing the error rate with the baseline for the old version.&lt;/li>&lt;li>Each invocation that failed on the new version should be retried on the old version, to ensure there&amp;rsquo;s no downtime (remember: you should have idempotency throughout).&lt;/li>&lt;li>A failed release (error rate increased beyond the acceptable) should be rolled back automatically, with alerts, so the development team can investigate.&lt;/li>&lt;li>For back-end applications, you&amp;rsquo;ll likely want a continuous deployment model, where each successfully built change is released to production and enters into effect immediately.&lt;/li>&lt;li>For web and mobile applications, you&amp;rsquo;ll want a similar deployment model, but the changes should be deployed with feature flags that make them inactive, to be activated at a later stage based on marketing campaigns, etc. For mobile, you should batch your changes, so that a new version is published to the app stores only once a week or so.&lt;/li>&lt;li>Data migrations should be performed as part of the release, by &lt;a href="https://kubernetes.io/docs/concepts/workloads/pods/init-containers/">Init Containers&lt;/a>. You should be able to control when your migrations run, so that some run before the new instances come up, some during the release, and some only after all the old instances have been brought down.&lt;/li>&lt;/ul>&lt;h2 id="logs">Logs&lt;/h2>&lt;ul>&lt;li>Each service should be able to log in plain statements, or in JSON, based on a configuration variable passed at runtime. The default should be JSON, overridden locally to be plain instead.&lt;/li>&lt;li>Each log entry should comply with a company-wide JSON schema. Services should be tested to check whether the log lines they produce comply with this schema, as part of the build pipelines.&lt;/li>&lt;li>The &lt;code>Invocation Context&lt;/code> should also have its own schema, included as an optional field in the log entry schema.&lt;/li>&lt;li>Logging long strings is incredibly slow in general, so the default minimum log level should be kept to &lt;code>INFO&lt;/code>, and &lt;code>INFO&lt;/code> should never be used as part of normal workflows.&lt;/li>&lt;li>An invocation should be able to specify a toggle to bump the produced logged level from &lt;code>DEBUG&lt;/code> to &lt;code>INFO&lt;/code>.&lt;/li>&lt;li>Does this mean that most workflows won&amp;rsquo;t produce any logs? No, because an event sink can map each produced event into a log line, asynchronously, without slowing down the processing itself.&lt;/li>&lt;li>So why do we need to occasionally bump the log level to &lt;code>INFO&lt;/code>? Or why log directly from the apps completely? That&amp;rsquo;s because, sometimes, weird things happen, and an application ends up not being able to publish events. Or you might be interested in how the inside of an app is working.&lt;/li>&lt;li>The events are already recorded forever, so no point in mapping them as logs that are also kept forever. You can keep the direct application logs for a while (1 month, or something similar), and merge these on demand with the event logs, when an employee wants to investigate something.&lt;/li>&lt;li>Imagine you&amp;rsquo;re having a live incident, you can ask your logging system to give you all the logs for a given invocation ID. Your system will merge the direct logs from the application containing that ID, and re-consume all events with that invocation ID, to produce a search space for you. Your system should show the whole tree of correlated logs and events.&lt;/li>&lt;li>If you create wrapping types for Personal Identifiable Information (PII) and secrets, you can easily avoid these being leaked in the direct application logs, and you can mask them when bringing them into the search space.&lt;/li>&lt;li>My advice is to use open-source tools for logs collection and searching, like &lt;a href="https://www.elastic.co/kibana">Kibana&lt;/a>, as commercial solutions can quickly get expensive.&lt;/li>&lt;/ul>&lt;h1 id="application-development">Application development&lt;/h1>&lt;h2 id="codebase">Codebase&lt;/h2>&lt;ul>&lt;li>You should choose whether you want a mono-repo or different repositories for your services.&lt;/li>&lt;li>Even when you use multiple programming languages (hopefully always), mono-repos make local development a lot easier, so this is something to consider.&lt;/li>&lt;li>If you use multiple languages and adopt a mono-repo, consider grouping your projects under language-specific folders. This is going to make your life easier when working with IDEs.&lt;/li>&lt;/ul>&lt;h2 id="frameworks-and-libraries">Frameworks and libraries&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to figure out how to cover a range of table stakes, including configuration parsing, etc. You don&amp;rsquo;t want to implement these low-level concerns yourself.&lt;/li>&lt;li>Choose libraries and frameworks that make your life easier, without becoming too attached to them, and avoiding the ones that are too opinionated or that significantly slow down application startup.&lt;/li>&lt;li>You should be able to check for new versions of every library a service uses, and to update them, with one command.&lt;/li>&lt;li>Each service should have a scheduled job to update all dependency versions (libraries, build tools, plugins, programming languages, frameworks, etc.), run the tests for that service, open a PR to update those dependencies if the tests pass, and automatically merge it. If the tests fail, the job should alert the development team.&lt;/li>&lt;/ul>&lt;h2 id="event-driven-optimizations">Event-driven optimizations&lt;/h2>&lt;ul>&lt;li>For event-processors, you&amp;rsquo;ll want to process an event on only one thread.&lt;/li>&lt;li>Ideally your processing should happen in-memory. Your service processes the inbound event, and update its in-memory state.&lt;/li>&lt;li>A separate process aggregates your outbound events into a state snapshot, using the partition as key, and including the offset the snapshot is valid at.&lt;/li>&lt;li>When your processor starts, it gets assigned a set of partitions, loads the state snapshot for each of them, and then compares the offset of the snapshot with the offset of the inbound message.&lt;/li>&lt;li>If the snapshot is behind, the processor simply reprocesses the events between the offset of the inbound message and the offset of the snapshot. The service applies the state changes to the in-memory state, and it&amp;rsquo;s back, ready to correctly process the inbound message it received when starting up.&lt;/li>&lt;li>For event sinks, you should adopt batched processing if the service is not on the critical path. As an example, a sink that updates a materialized view could batch every 5000 events or every 1 second, whichever happens first, and do 1 database batch insert instead of 5000 inserts and round-trips.&lt;/li>&lt;li>Perhaps unsurprisingly, in-memory processing hugely reduces latency while achieving great throughput, while batched processing enormously improves throughput.&lt;/li>&lt;/ul>&lt;h2 id="shared-libraries">Shared libraries&lt;/h2>&lt;ul>&lt;li>Some people swear by shared libraries, and some other folks swear when they see them. In any case, you should make an explicit decision about this.&lt;/li>&lt;li>Shared libraries are only available to programming languages compatible with the language they&amp;rsquo;re written in. So mixing multiple programming languages in the same area e.g., web apps, or back-end apps can be a pain.&lt;/li>&lt;li>Avoid home-made frameworks at all costs. Home-made libraries are okay. The difference between frameworks and libraries is that frameworks are heavily opinionated and require full buy-in, while libraries are modular and can be easily replaced.&lt;/li>&lt;li>If you decide to use shared libraries, each should be developed, versioned, built, tested, released, and documented like an external library.&lt;/li>&lt;li>Shared libraries can be powerful to offer well-thought plug-and-play behavior that would be expensive and error-prone to implement in each new service.&lt;/li>&lt;li>My advice is to write a set of shared libraries for each programming language you use.&lt;/li>&lt;li>In any case, never ever mandate any library or framework. Each team should decide what to use for each service, without being obligated to use a company-wide approach.&lt;/li>&lt;li>Enforcing consistency kills continuous improvement, because it increases the cost of choosing a different approach. Experiments become less appealing, and anything promising requires changing everybody&amp;rsquo;s opinion and updating all code. Don&amp;rsquo;t do this. Always assume you have two or more stable alternative approaches, and a few ongoing experiments. Every new initiative should involve an experiment.&lt;/li>&lt;/ul>&lt;h2 id="service-level-architecture">Service-level architecture&lt;/h2>&lt;ul>&lt;li>Adopt a standard service-level architecture, to organize the code that makes up each service for ease of readability and testability.&lt;/li>&lt;li>Organize your code so that what&amp;rsquo;s used together stays and changes together. So &amp;ldquo;winter clothes&amp;rdquo; and &amp;ldquo;summer clothes&amp;rdquo; are good groupings, while &amp;ldquo;trousers&amp;rdquo; and &amp;ldquo;tops&amp;rdquo; are bad.&lt;/li>&lt;li>My advice is to adopt the &lt;a href="https://alistair.cockburn.us/hexagonal-architecture/">Hexagonal Architecture&lt;/a>, also called Ports and Adapters Architecture. It&amp;rsquo;s a great way of structuring your code for effective changing and testing.&lt;/li>&lt;/ul>&lt;h2 id="id-generation">ID generation&lt;/h2>&lt;ul>&lt;li>You should choose the type of unique IDs you generate. Different types work best in different contexts.&lt;/li>&lt;li>My advice is to use &lt;a href="https://github.com/ulid/spec">Universally Unique Lexicographically Sortable Identifier&lt;/a>s (ULIDs) or &lt;a href="https://github.com/segmentio/ksuid">K-Sortable Unique Identifier&lt;/a>s (KSUIDs) for domain IDs, and &lt;a href="https://github.com/f4b6a3/tsid-creator">Time-Sorted Unique Identifier&lt;/a>s (TSIDs) as primary keys in SQL databases. Here&amp;rsquo;s &lt;a href="https://vladmihalcea.com/uuid-database-primary-key/">why TSIDs make great SQL database primary keys&lt;/a>.&lt;/li>&lt;li>You should also figure out whether to rely on sortable unique identifiers to infer the order of the events, in which case you should generate IDs from an event-processor, with partitioning, to ensure in-order processing.&lt;/li>&lt;/ul>&lt;h2 id="resilience">Resilience&lt;/h2>&lt;ul>&lt;li>Circuit breakers and retries are important whenever you perform requests to downstream services.&lt;/li>&lt;li>In an event-driven architecture, most services process messages from a queue, making both of these less important. For command and query endpoints, though, you still need them.&lt;/li>&lt;li>For blocking IO operations, you&amp;rsquo;ll want thread pools and bulkheads. An example might involve a separate thread pool and connection pool for each tenant, to interact with a SQL database. This way a tenant cannot easily influence other tenants with their activity.&lt;/li>&lt;li>Each service should also expose a set of management endpoints, for health checks, etc. It&amp;rsquo;s important that these are exposed on a separate port, rather than on the main application port, so that Kubernetes can access these without exposing them at all to other services or over public internet.&lt;/li>&lt;li>Another good practice is a separate thread pool for the management endpoints, for otherwise high activity will make your application not responsive to management queries and commands.&lt;/li>&lt;/ul>&lt;h2 id="data-optimizations">Data optimizations&lt;/h2>&lt;ul>&lt;li>You should use compression for client-to-server, server-to-client, and service-to-service communications.&lt;/li>&lt;li>Large attachments e.g., images and videos should be uploaded to object storage instead (e.g., &lt;a href="https://cloud.google.com/storage">Google Cloud Storage&lt;/a> or &lt;a href="https://aws.amazon.com/pm/serv-s3/">AWS S3&lt;/a>), and a pre-signed download URL should be sent instead, as a mechanism to obtain the attachment if needed.&lt;/li>&lt;li>For images and videos, you also want to leverage CDNs, to dynamically adjust their size based on the screen dimensions of the user, and to choose their quality based on the user&amp;rsquo;s internet connection speed.&lt;/li>&lt;li>Whenever a user needs to upload an attachment (image, video, etc.), your back-end should generate a pre-signed upload URL to an object storage location, so that the traffic doesn&amp;rsquo;t flow through your systems.&lt;/li>&lt;li>You should adopt a quarantine bucket, so that every upload triggers a malware inspection process, which copies the file to the actual bucket if it deems the attachment secure.&lt;/li>&lt;li>About PII, you should either encrypt the whole document with an end-user-specific key so, after the user has invoked the Right To Be Forgotten the documents will be unusable.&lt;/li>&lt;li>Alternatively, you should store a template for the document and keep the data separate, so that after the user has opted out, you would still be able to produce a document, but with some of the information redacted.&lt;/li>&lt;/ul>&lt;h2 id="dealing-with-money">Dealing with money&lt;/h2>&lt;ul>&lt;li>You should model money so that invalid amounts e.g., $15.732 cannot be represented.&lt;/li>&lt;li>To avoid representation precision issues with decimals, you should model currency amounts using the fundamental units of a given currency. So instead of representing $7.52 dollars with a decimal, you should represent it with the number of dollar cents, 752.&lt;/li>&lt;li>Tenants and users based in different countries might get billed in different currencies. Keep this in mind if you have this requirement, and think about currency conversion.&lt;/li>&lt;/ul>&lt;h2 id="encryption">Encryption&lt;/h2>&lt;ul>&lt;li>Choose a battle-tested library for your encryption primitives. Never ever roll your own implementation of anything encryption related (unless you&amp;rsquo;re an encryption company). &lt;a href="https://www.bouncycastle.org/">BouncyCastle&lt;/a> is a great choice.&lt;/li>&lt;li>Figure out if you need compliance with the &lt;a href="https://en.wikipedia.org/wiki/Federal_Information_Processing_Standards">Federal Information Processing Standards&lt;/a> (FIPS). Unless your work closely with the US government, you should probably stick to non-FIPS implementations, as they are better supported and arguably more secure. Here&amp;rsquo;s &lt;a href="https://techcommunity.microsoft.com/t5/microsoft-security-baselines/why-we-re-not-recommending-fips-mode-anymore/ba-p/701037">an explanation of why Microsoft stopped recommending FIPS-compliant implementations&lt;/a>.&lt;/li>&lt;li>Consider whether you would benefit more from runtime selection of algorithms and parameters, or from type-safe and abstracted encryption primitives. Personally I never found the first useful, and hate using BouncyCastle&amp;rsquo;s API directly.&lt;/li>&lt;li>Stick to well-known and battle-tested algorithms, like &lt;a href="https://en.wikipedia.org/wiki/Advanced_Encryption_Standard">AES&lt;/a> for symmetric encryption, and &lt;a href="https://en.wikipedia.org/wiki/Secure_Hash_Algorithms">SHA&lt;/a> or &lt;a href="https://en.wikipedia.org/wiki/Argon2">Argon2&lt;/a> for hashing (SHA and Argon2 are best for different use cases).&lt;/li>&lt;li>If your application relies on asymmetric encryption at application level e.g., a distributed ledger, you might want to consider post-Quantum algorithms like &lt;a href="https://pq-crystals.org/kyber/">Crystals Kyber&lt;/a> (as key encapsulation mechanism or KEM) and &lt;a href="https://pq-crystals.org/dilithium/">Crystals Dilithium&lt;/a> (as a digital signature scheme).&lt;/li>&lt;/ul>&lt;h1 id="building-and-testing">Building and testing&lt;/h1>&lt;h2 id="packaging">Packaging&lt;/h2>&lt;ul>&lt;li>Each service should be packaged, as part of its build. You should choose between &lt;a href="https://github.com/opencontainers/image-spec">OCI images&lt;/a> e.g., &lt;a href="https://www.docker.com/">Docker&lt;/a>, or native images e.g., &lt;a href="https://www.graalvm.org/">Graal&lt;/a>.&lt;/li>&lt;li>You should also tweak the runtime properties for your executable bundles. For JVM-based applications, this means chosen garbage collector, memory settings, GC flags, additional JVM options, etc.&lt;/li>&lt;li>Structure your OCI images with &lt;a href="https://docs.docker.com/build/building/multi-stage/">multi-stage builds&lt;/a>, so that each stage can be cached if it didn&amp;rsquo;t change.&lt;/li>&lt;li>Ensure you separate runtime requirements from build-time requirements, so unnecessary things don&amp;rsquo;t end up in your image.&lt;/li>&lt;li>Your &lt;a href="https://medium.com/nttlabs/bit-for-bit-reproducible-builds-with-dockerfile-7cc2b9faed9f">OCI images should be reproducible&lt;/a>, meaning that the same code version should produce identical bytes. This requires some configuration tuning, but also for the contained artifacts e.g., JARs to also be reproducible. Here&amp;rsquo;s &lt;a href="https://docs.gradle.org/current/userguide/working_with_files.html#sec:reproducible_archives">how to enable reproducible archives in Gradle&lt;/a>.&lt;/li>&lt;/ul>&lt;h2 id="builds">Builds&lt;/h2>&lt;ul>&lt;li>Choose a smart build tool, so that only modules whose hash has changed get rebuilt and re-tested. This can save a ton of time from the duration of your builds. &lt;a href="https://gradle.org/">Gradle&lt;/a> does this well.&lt;/li>&lt;li>Write scripts to build each service locally, without having to know the details of the build tool used by that service.&lt;/li>&lt;li>Embed the build tool within the codebase of the service, rather than expecting for this to be installed on the machine running the build. An example is the &lt;a href="https://docs.gradle.org/current/userguide/gradle_wrapper.html">Gradle Wrapper&lt;/a>.&lt;/li>&lt;li>Your build pipeline should be the only criteria to decide whether your service is releasable. No human inspections should be needed. Manual regression testing is a crime. Manual exploratory testing is great, but not as part of the build and release pipeline.&lt;/li>&lt;li>You should be able to build all your systems locally, relying only on localhost, and expecting nothing to be installed on the machine.&lt;/li>&lt;li>As an example, if Java is not installed but is needed to build a service, the build script should install Java. An alternative approach to achieve this is to use a Docker container to run the build itself.&lt;/li>&lt;li>Decide whether you prefer &lt;a href="https://github.com/features/actions">GitHub Actions&lt;/a>, or a traditional build server e.g., &lt;a href="https://www.jetbrains.com/teamcity/">TeamCity&lt;/a>. Build servers make it easy to rebuild dependent services, etc. and are generally much faster.&lt;/li>&lt;li>If you prefer GitHub Actions, seriously consider running the agent on your own Kubernetes cluster, so that each build runs in its own throwaway Docker container on Kubernetes. It&amp;rsquo;ll be much faster and cheaper.&lt;/li>&lt;li>Ensure you upload build and test reports, and that you alert the development team of any failed builds, and of any build that was previously failing and just succeeded.&lt;/li>&lt;li>Each service should specify its own build script, producing a Docker image, and uploading it to an image repository.&lt;/li>&lt;li>Tag each image with the full hash of the code from version control, and pass this as an argument to the image, so that the service will be able to return its version from a management endpoint.&lt;/li>&lt;li>On the remote build pipeline, sign each image with a key-pair, tag the image with the signature and the key ID, and verify the signature against the relevant public key before allowing Kubernetes to spin up those images. This prevents people from uploading imagines not produced by your build pipelines, whether unintentionally or maliciously.&lt;/li>&lt;/ul>&lt;h2 id="functional-tests-as-part-of-the-build">Functional tests as part of the build&lt;/h2>&lt;ul>&lt;li>Each service should declare its input and output contract, using &lt;a href="https://www.openapis.org/">OpenAPI&lt;/a> specifications, Avro schemata, JSON schemata, etc.&lt;/li>&lt;li>Test each service against its input and output contracts, to see whether it stays compatible with other services.&lt;/li>&lt;li>Test each declared input and output contract for compliance with a set of company-wide rules. Examples include path style for OpenApi, or kebab-case instead of camelCase for Avro and JSON. This guarantees you get consistency of API standards, and is especially important for customer-facing contracts.&lt;/li>&lt;li>Test each internal module against its code contract (the externally visible types and functions). Do this wrong, and your tests will be tied to the internal structure of your code, meaning they&amp;rsquo;ll break when you change the structure, even if the behavior still works.&lt;/li>&lt;li>Test each driving and driven adapter against the technology they integrate with. These integration tests should run using only localhost. Use a library like &lt;a href="https://testcontainers.com/">Testcontainers&lt;/a> to orchestrate Docker containers as part of your tests.&lt;/li>&lt;li>After each driving and driven adapter is tested in isolation, and your business logic is covered by the contract tests for the application and domain modules, have one or two service tests. These start the whole service locally, and go from a driving adapter to a driven adapter, testing the correct configuration and wiring of the service.&lt;/li>&lt;li>Run your service tests against your service running in-memory, and against the same service running as a Docker image. This way, you can identify issues with the packaging part.&lt;/li>&lt;li>Use &lt;a href="https://pitest.org/">mutation testing&lt;/a> to check whether your tests are actually protecting you against undesired code changes. Aim for zero survived mutants. Don&amp;rsquo;t bother with measuring test code coverage.&lt;/li>&lt;li>Avoid any dependencies among tests, whether it&amp;rsquo;s shared data or expected order of execution.&lt;/li>&lt;li>The whole build, including all tests and checks should run in under 3 to 5 minutes, and you should be able to run all tests and checks locally. Having a smart build tool like Gradle helps with this.&lt;/li>&lt;/ul>&lt;h2 id="checks-as-part-of-the-build">Checks as part of the build&lt;/h2>&lt;ul>&lt;li>Perform static code quality checks.&lt;/li>&lt;li>Automatically apply reformatting and linting.&lt;/li>&lt;li>Inspect the code to prevent accidental leakage of secrets.&lt;/li>&lt;li>Perform static code security analysis.&lt;/li>&lt;li>Check dependency versions against known vulnerabilities.&lt;/li>&lt;li>The whole build, including all tests and checks should run in under 3 to 5 minutes, and you should be able to run all tests and checks locally. Having a smart build tool like Gradle helps with this.&lt;/li>&lt;/ul>&lt;h2 id="performance-tests-as-part-of-the-build">Performance tests as part of the build&lt;/h2>&lt;ul>&lt;li>For parts of your codebase where performance is critical, create micro-benchmarks e.g., with &lt;a href="https://github.com/openjdk/jmh">JMH&lt;/a> if you&amp;rsquo;re using a JVM-based language.&lt;/li>&lt;li>Establish a baseline, and run these as part of each build, failing the process if there are regressions in the measured performance.&lt;/li>&lt;li>For event-driven workflows, you can use flow analysis to determine the performance of the whole system from the performance of the parts. This is an extremely nice property. Given an event processor, you should test the maximum throughput it can sustain, and the latency each event experiences.&lt;/li>&lt;li>The whole build, including all tests and checks should run in under 3 to 5 minutes, and you should be able to run all tests and checks locally. Having a smart build tool like Gradle helps with this.&lt;/li>&lt;/ul>&lt;h2 id="smoke-tests-in-production">Smoke tests in production&lt;/h2>&lt;ul>&lt;li>You should have a minimal set of smoke tests, running periodically in production.&lt;/li>&lt;li>These should not duplicate the tests you run as part of each build, and you&amp;rsquo;re not aiming for coverage here.&lt;/li>&lt;li>The only purpose of smoke tests is to catch unexpected problems in your production environment. Misconfigured network rules, etc.&lt;/li>&lt;li>Having one test for each main workflow works well, and these tests should exercise your whole system, in depth.&lt;/li>&lt;li>An example of a smoke test might involve creating a test user for a test tenant, failing to log in with the wrong password, resetting the password by following a link on an email received by a test email server, logging in successfully, logging out, and checking that a login-protected page is now not available.&lt;/li>&lt;li>If a smoke test fails, something is seriously wrong.&lt;/li>&lt;li>You&amp;rsquo;ll need test tenants, test users, test email servers.&lt;/li>&lt;li>Normally you should use your actual third-party dependencies. Sometimes this is not possible, at which point you should use their test versions. If your 3rd-party services don&amp;rsquo;t have test versions, replace them with stubs configured to only be enabled for these test users.&lt;/li>&lt;li>These smoke tests should run in production and nowhere else, because the problems these tests catch are almost always environment-specific, so running them in another environment would be completely useless.&lt;/li>&lt;li>Ideally these should run in under 20 minutes, and you should run them every 30 minutes or so.&lt;/li>&lt;/ul>&lt;h2 id="resilience-tests-in-production">Resilience tests in production&lt;/h2>&lt;ul>&lt;li>You should perform resilience tests in your production environment. &lt;a href="https://en.wikipedia.org/wiki/Chaos_engineering">Chaos engineering&lt;/a> tools are a good approach to achieve this.&lt;/li>&lt;li>You should resist the temptation to run these in a separate environment, as they&amp;rsquo;d prove very little.&lt;/li>&lt;/ul>&lt;h2 id="runtime-monitoring-in-production">Runtime monitoring in production&lt;/h2>&lt;ul>&lt;li>Each event-driven workflow should produce an alert if a step doesn&amp;rsquo;t complete within a time limit, or if the whole workflow exceeds a time limit.&lt;/li>&lt;li>You should have a runtime security agent checking for abnormal behavior happening in production. &lt;a href="https://go.wiz.io/">Wiz&lt;/a> and &lt;a href="https://www.aquasec.com/">Aquasec&lt;/a> can help with this.&lt;/li>&lt;/ul>&lt;h2 id="exploratory-performance-tests">Exploratory performance tests&lt;/h2>&lt;ul>&lt;li>Exploratory performance tests should be run in ephemeral copies of your production environment.&lt;/li>&lt;li>You should be careful with these, as the amount of data you have will likely influence your performance, so either make the two environments identical or be careful about what you can infer from your tests.&lt;/li>&lt;li>Examples of these tests include &lt;a href="https://zoonou.com/resources/blog/performance-testing-techniques/">load tests, stress tests, spike tests, and soak tests&lt;/a>.&lt;/li>&lt;/ul>&lt;h1 id="data-and-analytics">Data and analytics&lt;/h1>&lt;h2 id="data-lake">Data lake&lt;/h2>&lt;ul>&lt;li>Every application event should include all the relevant and available information known at the time, without leaving anything out because it&amp;rsquo;s not currently used by any downstream event processor.&lt;/li>&lt;li>Never run &lt;a href="https://en.wikipedia.org/wiki/Online_analytical_processing">OLAP operations&lt;/a> on a database used for &lt;a href="https://en.wikipedia.org/wiki/Online_transaction_processing">OLTP operations&lt;/a> by a service. Instead, an event sink can map the domain events published to a data lake reserved for OLAP queries.&lt;/li>&lt;li>Ensure you mask all sensitive data before inserting your events&amp;rsquo; information into the data lake.&lt;/li>&lt;/ul>&lt;h2 id="materialized-aggregations">Materialized aggregations&lt;/h2>&lt;ul>&lt;li>You can use event sinks to update a materialized view every time the information it provides a snapshot of changed. This yields analytics that are updated as soon as the event that changed them is processed. An example would be the total count of users, updated every time an event indicates that a new user has joined.&lt;/li>&lt;/ul>&lt;h2 id="data-operations-and-custom-dashboards">Data operations and custom dashboards&lt;/h2>&lt;ul>&lt;li>Use a set of connectors to produce events about various aspects. These should not include application events, as those are already produced by your applications.&lt;/li>&lt;li>Anything of interest that happens off-platform should be produced. Financial reports, etc.&lt;/li>&lt;li>Adopt tools that allow every employee to create their own custom dashboards and metrics. An example of these tools is &lt;a href="https://www.metabase.com/">Metabase&lt;/a>.&lt;/li>&lt;/ul>&lt;h1 id="back-office">Back-office&lt;/h1>&lt;h2 id="internal-application">Internal application&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need a back-office web application for your employees.&lt;/li>&lt;li>This should be built with the same care of your main web applications.&lt;/li>&lt;li>Every operation should be event-driven, with full auditability.&lt;/li>&lt;li>You might want to support different roles, and require admin review and approval for some operations.&lt;/li>&lt;li>This web application shouldn&amp;rsquo;t be accessible over public internet, but only through a &lt;a href="https://en.wikipedia.org/wiki/Virtual_private_network">Virtual private Network&lt;/a>.&lt;/li>&lt;li>This back-office application should use federated access with your company&amp;rsquo;s Identity Provider (IDP), so that your employees can log in to it with Single Sign-On, and with the appropriate authorization roles.&lt;/li>&lt;/ul>&lt;h2 id="tenant-management">Tenant management&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to list your tenants, and manage their configuration.&lt;/li>&lt;li>You should be able to see which features and modules they have enabled, the weights and limits they have, and to change these settings.&lt;/li>&lt;li>You need to be able to onboard a tenant from here, unless they can onboard from the app itself, in which case you need notifications.&lt;/li>&lt;li>You should be able to review any applications, and either approve them or reject them for a specific reason. This should cause notifications for the tenant, received in-app and over email.&lt;/li>&lt;/ul>&lt;h2 id="user-management">User management&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to list and manage your users.&lt;/li>&lt;li>You should be able to browse a user&amp;rsquo;s activity.&lt;/li>&lt;li>Sensitive information (including names, phone numbers, email addresses, transaction details, etc.) should be shown encrypted. The decision to decrypt should stay on record for audit purposes.&lt;/li>&lt;/ul>&lt;h2 id="issue-management">Issue management&lt;/h2>&lt;ul>&lt;li>You&amp;rsquo;ll need to list and manage the issues your users have reported from your products.&lt;/li>&lt;li>You should be notified whenever a user reports an issue.&lt;/li>&lt;li>Issues should be presented grouped by time, category, similarity, user, etc.&lt;/li>&lt;li>From a list view, you should be able to focus on an issue, check the user, see the actions leading to the issue, find all events and logs related to the invocation ID, and make this available to some developers for further investigation.&lt;/li>&lt;li>You should also be able to perform specific actions to compensate the user, or to fix their state. These should be normal commands, captured as events, showing up in the user&amp;rsquo;s history, etc.&lt;/li>&lt;li>When done, you should be capable of notifying the user in-app about the fact that the issue is resolved, and to explain what the cause was. This needs to be reflected on the issue itself.&lt;/li>&lt;li>Whenever things are more involved, you should be able to contact the user that&amp;rsquo;s having the issue through the in-app chat.&lt;/li>&lt;li>A back-office user should be able to template a resolution, and to then apply it to multiple related issues.&lt;/li>&lt;li>Back-office operators should have dashboards showing activity levels and error rates, with comparisons with historical averages.&lt;/li>&lt;/ul>&lt;h2 id="product-management">Product management&lt;/h2>&lt;ul>&lt;li>Product operators should be able to manage features, modules, products, subscription tiers (with their limits, descriptions, etc.), product requirements, in-app marketing messages, etc.&lt;/li>&lt;li>You should be able to plan the release of a feature, with scheduling, marketing message, incremental rollout, etc.&lt;/li>&lt;li>Develop the ability to A/B test features, by enabling a toggle for a percentage of invocations for a target users sample (based on characteristics, etc.)&lt;/li>&lt;li>Be able to browse product activity, with trends, patterns, etc. Also, operators should be able to deep dive into specific user sessions, to understand users&amp;rsquo; behavior.&lt;/li>&lt;li>Operators should be able to manage dashboards and derived metrics.&lt;/li>&lt;/ul>&lt;h2 id="infrastructure-management">Infrastructure management&lt;/h2>&lt;ul>&lt;li>Infrastructure operators should be able to list infrastructure deployments, along with which tenants they host.&lt;/li>&lt;li>Develop the ability to show topics, event schemata, messaging queues, and services, in a topology.&lt;/li>&lt;li>Also, support visualizing your information flows, in terms of which services produce and consume which topics, laid out like a graph. Include a quantitative visualization of your traffic patterns: how many events are moving, on which topics, the size of your queues, etc.&lt;/li>&lt;li>Give your operators the ability to read specific service information (by embedding your service registry or by linking to it).&lt;/li>&lt;li>Allow infrastructure operators to propose a scripted database operation, opening a case for somebody to review and approve, so that the infrastructure can then pick it up, execute it, and give you the result back. This should stay on record, to achieve auditability of database changes.&lt;/li>&lt;li>Enabled your infrastructure operators to query logs and events, in an auditable and controlled way.&lt;/li>&lt;/ul></description></item><item><title>Organizational principles for effective software product development</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-11-16-organizational-principles-for-effectiveness/</link><pubDate>Thu, 16 Nov 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-11-16-organizational-principles-for-effectiveness/</guid><description>&lt;p>I often rant about what I feel is wrong with an approach, practice, or system. Today, instead, I want to describe some principles that an effective software product development company should adopt.&lt;/p>&lt;p>Why principles? Wouldn&amp;rsquo;t a detailed description of how things should work be more useful? It&amp;rsquo;d be almost impossible to write, as it&amp;rsquo;d be a colossal task. Perhaps it&amp;rsquo;d be good for a book, but definitely not for a blog post.&lt;/p>&lt;p>Also, reality is multifaceted and nuanced. Everything depends on everything. Prescriptive approaches don&amp;rsquo;t cope well with dynamic situations. Principles, on the other hand, allow you to deal with uncertainty.&lt;/p>&lt;h1 id="how-to-use-these-principles">How to use these principles&lt;/h1>&lt;p>You can use these principles to design how a company operates. While you&amp;rsquo;re designing, check whether the system violates any of the principles. If it does, design an alternative one that doesn&amp;rsquo;t.&lt;/p>&lt;p>These principles work in synergy, so the value they bring when implemented together is more than the sum of the parts. However, they&amp;rsquo;re also useful on their own, which means that you can apply them when designing not only whole organizations, but also departments, areas, or teams.&lt;/p>&lt;p>At the very least, they provide a framework to evaluate how effectively a company or a team operates.&lt;/p>&lt;h1 id="applicability">Applicability&lt;/h1>&lt;p>Like any model, these principles have specific applicability boundaries. They&amp;rsquo;re only useful when:&lt;/p>&lt;ol>&lt;li>There&amp;rsquo;s a willingness to strive for excellence. If a company is content with merely doing okay, some of these will sound extreme or dangerous.&lt;/li>&lt;li>The results are placed before what&amp;rsquo;s convenient for some people. Keeping people with their comfort zone, supporting the existing control structures, and feeding people&amp;rsquo;s ego are non-goals.&lt;/li>&lt;/ol>&lt;p>This list is also incomplete. So if I don&amp;rsquo;t mention a specific good principle you have in mind, it doesn&amp;rsquo;t mean I&amp;rsquo;m saying it&amp;rsquo;s not important.&lt;/p>&lt;h1 id="principles">Principles&lt;/h1>&lt;h2 id="outside-in-structure">Outside-in structure&lt;/h2>&lt;p>Employees work as part of different levels. Activities range from being closest to the customers, to the innermost level. The time horizons tend to widen as the levels get further from the customers.&lt;/p>&lt;p>The inner levels support the outer levels. It&amp;rsquo;s about coordination, not command and control. The outer levels can pull help from the inner levels, but the decision-making stays with the people working as close as possible with the customers.&lt;/p>&lt;p>The people that work in the inner levels tend to manage how the work works, and shepherd long-term goals, initiatives, and changes.&lt;/p>&lt;p>The opposite of this model is the traditional top-down structure, where each level controls the level below it.&lt;/p>&lt;h2 id="the-unit-of-work-allocation-is-a-team">The unit of work allocation is a team&lt;/h2>&lt;p>Work is carried over by teams, rather than by individuals. This means that there are no individual tasks or assignments.&lt;/p>&lt;p>A team is a group of people working together towards the same goal. Sharing a goal is not enough, people need to collaborate synchronously towards achieving it. Ensemble programming is a great example of a practice that embodies this principle.&lt;/p>&lt;h2 id="small-equivalent-teams">Small equivalent teams&lt;/h2>&lt;p>Teams should be as small as possible, while still qualifying as a team. A team must abstract who&amp;rsquo;s doing what, be capable of doing any work, and be resilient to absences. To achieve this, teams should be cross-functional and staff people with different backgrounds and strengths.&lt;/p>&lt;p>You want a mix of old and young, senior and junior, academic and self-taught, genders, etc. You also need shapers, fixers, and explorers. A team should also include both starters and finishers. About skills, you should at least cover web development, user experience, distributed systems, codebase structure, Cloud infrastructure, databases, product discovery, scripting, observability, and code maintainability.&lt;/p>&lt;p>People are not interchangeable, and the same person might complement a team while not being a good fit in others. Hire for the specific skills, strengths, archetype, personality, and background you need in a given team.&lt;/p>&lt;p>I believe that 5 team members is the sweet spot in most cases. You can get the necessary diversity, are resilient to 1 or 2 absences, and you don&amp;rsquo;t get so many people that they inevitably need to work on more than 1 thing at a time.&lt;/p>&lt;p>The design goal here is to eliminate cross-team dependencies.&lt;/p>&lt;h2 id="work-in-progress-limits">Work-in-progress limits&lt;/h2>&lt;p>Each team should be working on only 1 initiative at a time. So the number of teams you have is the degree of parallelism you get. Not everything counts as an item though, only high-level product goals or technical initiatives do.&lt;/p>&lt;p>A team is always improving various things, but only working on 1 significant piece of work at a time.&lt;/p>&lt;h2 id="pull-based-just-in-time-work">Pull-based just-in-time work&lt;/h2>&lt;p>The teams should pull work when they finish their previous work. As a team finishes working, they should pull the next initiative or goal and start working on it.&lt;/p>&lt;p>This means that teams cannot get overwhelmed by having work pushed on them. Information about upcoming work shouldn&amp;rsquo;t reach a team until that team has finished their current initiative.&lt;/p>&lt;h2 id="pull-based-information">Pull-based information&lt;/h2>&lt;p>Documentation, expertise, advice, support, and other types of information should also be pull-based, and made available when somebody needs it.&lt;/p>&lt;h2 id="leaderless-teams">Leaderless teams&lt;/h2>&lt;p>A team should be a group of equals who are great at different things. Nobody should be the designated leader of the team. Seniority shouldn&amp;rsquo;t formally matter. The team can self-organize in terms of representation when dealing with stakeholders, but nobody should be allowed to represent the team formally or indefinitely.&lt;/p>&lt;h2 id="company-wide-initiative-queues">Company-wide initiative queues&lt;/h2>&lt;p>A company should organize and prioritize goals and initiatives using company-wide queues. This means that prioritization should happen across areas, rather than within each area.&lt;/p>&lt;p>This works well with equivalent teams capable of working on anything, so that the company can focus on whatever is needed.&lt;/p>&lt;p>This is in opposition to deciding what&amp;rsquo;s next for each team or area, leading to inflexible work allocation and suboptimal ability to focus.&lt;/p>&lt;h2 id="direct-responsible-individuals-for-initiatives">Direct Responsible Individuals for initiatives&lt;/h2>&lt;p>Each initiative should have a Direct Responsible Individual (DRI), who&amp;rsquo;s in charge of coordinating the work, escalating risks and issues, and outlining dependencies. The DRI is not responsible for the outcomes, and is also not a decision-maker.&lt;/p>&lt;p>When decisions are of medium importance, the DRI might make a call, but anybody can decide to escalate the decision to people who work in inner levels.&lt;/p>&lt;p>Being a DRI is not a job, and it&amp;rsquo;s always a temporary engagement. The rule is that if an initiative you were DRI for just ended, you&amp;rsquo;re not going to be DRI in the next initiative you work on. The DRI is nominated autonomously by the team that commits to an initiative.&lt;/p>&lt;h2 id="tribes-as-smaller-companies">Tribes as smaller companies&lt;/h2>&lt;p>A company should behave as a whole, until it reaches the critical number of 50 developers, roughly 10 teams. At that moment, the company should adopt tribes to coordinate the work better.&lt;/p>&lt;p>Tribes are like mini-companies within the wider company. The principles apply in the same way, except for the fact that everything stays within a tribe (teams, initiative queues, prioritization, etc.). Tribes should focus on end-to-end vertical slices of the business, not on a given function.&lt;/p>&lt;p>When a tribe becomes too big, and has 10 or more teams, it should create nested tribes within itself. Each nested tribe relates to the parent tribe in the same way the parent tribe relates to the company.&lt;/p>&lt;p>The design goal here is to eliminate cross-tribe dependencies.&lt;/p>&lt;h2 id="people-who-work-together-sit-close-to-each-other">People who work together sit close to each other&lt;/h2>&lt;p>People in a team should sit together. Teams that belong to the same tribe should also be near each other. Experts should go to the team that needs their expertise, and be there for the duration of the engagement.&lt;/p>&lt;h2 id="evolutionary-approach-to-development">Evolutionary approach to development&lt;/h2>&lt;p>Important decisions and initiatives should be de-risked by leveraging a set-based approach. Different teams should work in parallel towards the same goal, or on the same initiative, without any contact between them, except checking they&amp;rsquo;re not converging towards the same solution.&lt;/p>&lt;p>The importance of the initiative determines the number of teams working on it in parallel. At the end, the different solutions produced can be compared across multiple dimension, and a higher-quality decision or solution is reached.&lt;/p>&lt;p>This is a frequent practice in collaborative knowledge work (policymaking, city planning, vaccine development, scientific research, etc.), and in opposition to the standardization and efficiency factories obsess about.&lt;/p>&lt;h2 id="vertically-aligned-people">Vertically-aligned people&lt;/h2>&lt;p>People should be associated with their vertical product organization, their tribe, as opposed to their horizontal function-specific one. As an example, a Java back-end engineer in the &amp;ldquo;Everyday Spending&amp;rdquo; tribe would be a product developer within the &amp;ldquo;Everyday Spending&amp;rdquo; tribe, rather than a Java developer, or a back-end engineer.&lt;/p>&lt;p>Horizontal structures can exist, but none can be across the vertical structures, only contained within them. So, if the &amp;ldquo;Savings&amp;rdquo; tribe decides to adopt a &amp;ldquo;back-end&amp;rdquo; chapter to try and coordinate the continuous learning in that area, they&amp;rsquo;re free to do so, but that organization only lives within that tribe, never across tribes.&lt;/p>&lt;p>This means that anybody in a tribe is aligned to the cross-functional lead for that tribe, regardless of their own area of expertise. So if a lawyer works within the &amp;ldquo;Savings&amp;rdquo; tribe, they work under the remit of the &amp;ldquo;Head of Savings&amp;rdquo;, not of the &amp;ldquo;Head of Legal&amp;rdquo;.&lt;/p>&lt;h2 id="collective-code-ownership">Collective code ownership&lt;/h2>&lt;p>The teams collectively own the whole codebase of the company. No team or person owns any part of the codebase. Nobody needs to ask permission to anybody else to access and modify any code.&lt;/p>&lt;p>This holds true even across tribes. It&amp;rsquo;s natural that some teams and tribes will focus more on specific areas, but this changes nothing about this point.&lt;/p>&lt;h2 id="collective-product-ownership">Collective product ownership&lt;/h2>&lt;p>The teams collectively own the product and the features. No team or person owns an area of the product.&lt;/p>&lt;p>This holds true even across tribes. It&amp;rsquo;s natural that the tribes will tend to work on separate areas of the product, and this is encouraged, but a tribe should never have to ask permission to another tribe to implement or modify a feature.&lt;/p>&lt;h2 id="continuous-collaboration">Continuous collaboration&lt;/h2>&lt;p>The teams continuously collaborate, both within the team and with sponsors, experts, managers, cross-functional leads, and function leads. The preferred way of collaboration is synchronous. Whoever liaises with a team joins that team for the duration of the engagement.&lt;/p>&lt;p>From first principles, fully remote teams can hardly ever be as effective as physically co-located teams. It does not mean that running a fully remote team is not feasible, but the collaboration, even when done perfectly, will inherently be less effective than a perfectly-structured in-person equivalent. This might be compensated by other factors. Nonetheless, physically co-located teams remain the preference. If a company prefers to work remotely, a team should enlist members that are geographically near to each other in terms of timezone. Hybrid teams should be avoided.&lt;/p>&lt;p>When a team commits to work on an initiative, they engage with that initiative&amp;rsquo;s sponsor. This close collaboration lasts for the entire duration of the work, and terminates when the team pulls another initiative that is sponsored by another person.&lt;/p>&lt;h2 id="product-managers-stay-with-the-value-stream">Product managers stay with the value stream&lt;/h2>&lt;p>Product management experts are not part of the teams, but focus on a value stream. Each value stream can have one or more dedicated product managers. A product manager can coordinate the work for more than 1 value stream, even if they&amp;rsquo;re ideally single-threaded.&lt;/p>&lt;p>Every time a product initiative is part of a value stream, that value stream&amp;rsquo;s PM is the sponsor for the initiative. The team working on the initiative will work closely with them for the duration of the engagement. The teams remain responsible for the decision-making, while the sponsor acts as stakeholder and advisor.&lt;/p>&lt;h2 id="experts-stay-within-the-technical-value-stream">Experts stay within the technical value stream&lt;/h2>&lt;p>Experts coordinate the work for their technical value streams. A security expert will act as sponsor for all the technical initiatives part of the security value stream. In terms of collaboration with the teams, they behave like product managers do. As usual, the teams remain responsible for the decision-making, while the sponsor acts as stakeholder and advisor.&lt;/p>&lt;h2 id="experts-are-available-as-advisors">Experts are available as advisors&lt;/h2>&lt;p>Experts are always available for advice, even when a team is working on an initiative outside of that person&amp;rsquo;s (technical) value stream. If a team needs help, the team calls in the expert, who joins the team for as long as the team deems necessary.&lt;/p>&lt;p>This applies to experts of all sorts, including architecture, product discovery, product observability, security, marketing, compliance, legal, etc.&lt;/p>&lt;h2 id="the-teams-own-the-way-they-work">The teams own the way they work&lt;/h2>&lt;p>The teams are responsible for the way they work. This includes hiring, firing (more on this below), onboarding, training, and the processes and tools the team uses. Managers can coordinate processes and activities to facilitate the adoption of what&amp;rsquo;s working well in specific teams, but the teams remain responsible for how they work.&lt;/p>&lt;p>The only constraint is that any way the team decides to work should respect these principles.&lt;/p>&lt;p>If a team routinely misbehaves or causes trouble, a mechanism (more on this below) is eventually triggered, and the team is disbanded.&lt;/p>&lt;h2 id="no-meetings">No meetings&lt;/h2>&lt;p>Meetings should be avoided and replaced by continuous synchronous collaboration. Cross-functional ensemble programming is a good way of avoiding meetings. When somebody wants to know something from a team, they join the team&amp;rsquo;s ensemble and have a chat.&lt;/p>&lt;p>The design goal is that every team member should have access to all the information.&lt;/p>&lt;h2 id="nobody-manages-anybody">Nobody manages anybody&lt;/h2>&lt;p>Somebody ultimately must be responsible for identifying and addressing bad actors and under-performers (more about this below), but nobody should manage anybody.&lt;/p>&lt;p>Nobody approves anybody&amp;rsquo;s holidays, and nobody approves purchases and expenses either. Smart credit cards, e.g., Pleo should be used to ensure control and visibility. Anybody should be able to stand in front of their team or the founders and explain why an expense was in the best interest of the company. Not being able to justify it should result in termination of employment.&lt;/p>&lt;p>The design goal here is freedom and responsibility.&lt;/p>&lt;h2 id="no-seniority-designations-specific-titles-or-promotions">No seniority designations, specific titles, or promotions&lt;/h2>&lt;p>Everybody working in the outer level, close to the customers, should be a &amp;ldquo;product developer&amp;rdquo;. Titles can reflect somebody&amp;rsquo;s tribe, but this should have no impact in terms of compensation, benefits, or treatment. A title might look like &amp;ldquo;Product developer, Everyday Spending&amp;rdquo;.&lt;/p>&lt;p>Seniority designations should also not appear in any form. No senior, lead, staff, or principle anything. Promotions should also not happen. People should only change role when they want to, and when it&amp;rsquo;s needed by the company. In any case, compensation should exclusively be a function of the years of experience the person has, their main skills, their location, and their tenure within the company.&lt;/p>&lt;p>This is in opposition to career ladders and progression frameworks.&lt;/p>&lt;h2 id="profit-sharing-schemes">Profit-sharing schemes&lt;/h2>&lt;p>No person or team should receive a bonus. Individual or team bonuses foster internal competition, and that&amp;rsquo;s to be avoided.&lt;/p>&lt;p>The whole company should participate in a profit-sharing scheme, distributed according to a person&amp;rsquo;s years of experience, primary skills, and tenure in the company.&lt;/p>&lt;p>This is in opposition to individual performance-related bonuses.&lt;/p>&lt;h2 id="no-performance-reviews">No performance reviews&lt;/h2>&lt;p>No performance reviews should occur, for neither teams nor individuals. There should be mechanisms to address bad actors and under-performance, as described below, but these are triggered as soon as the problem is spotted, not quarterly or at the end of the year.&lt;/p>&lt;p>People are not compared, as they&amp;rsquo;re not equivalent. Teams might occasionally be compared, with the only goal of collectively improving the ways of working across the teams.&lt;/p>&lt;h2 id="continuous-integration">Continuous Integration&lt;/h2>&lt;p>No other branch that the main branch should ever be used. Each team should push small increments directly to the remote main branch, after running all tests locally. Whenever the main branch gets broken, the team should stop what they&amp;rsquo;re doing and fix it forward, rather than rolling the changes back.&lt;/p>&lt;p>Locally, the code should go from working state to working state, incrementally and iteratively, rather than be allowed to become broken before getting eventually fixed.&lt;/p>&lt;h2 id="continuous-delivery-even-on-fridays">Continuous Delivery (even on Fridays)&lt;/h2>&lt;p>Deployments should happen continuously, every day, without coordination, in an uneventful way. Each commit that&amp;rsquo;s pushed to the main branch should be automatically built, tested, packaged, inspected, and deployed. People shouldn&amp;rsquo;t even be aware of when the code is being deployed. Changes that require sales or marketing work should be enabled with feature flags after they were already released in the dark, with all the communications and campaigns they might need.&lt;/p>&lt;p>Avoiding deployments after a specific hour, or on specific days, e.g., on Fridays is a red flag. Any coordinated release approach e.g., code freezes, tags, or &amp;ldquo;keeping an eye on it&amp;rdquo;, is also a red flag.&lt;/p>&lt;h2 id="the-only-manual-testing-is-exploratory">The only manual testing is exploratory&lt;/h2>&lt;p>No regression testing should ever be performed manually. Manual testing should be strictly exploratory. Teams should be encouraged to routinely perform exploratory testing.&lt;/p>&lt;p>A cross-functional team dedicated to exploratory testing isn&amp;rsquo;t a bad idea, but exploratory testing should leave outside the software product lifecycle. In particular, manual testing as part of the release process is a red flag.&lt;/p>&lt;h2 id="only-production">Only production&lt;/h2>&lt;p>Production should be the only permanent environment. People should use internal tenants and users for exploratory testing, smoke tests, and live demos, directly in production.&lt;/p>&lt;p>Changes so dangerous than they are impossible to feature-flag or rolled out using a blue/green approach should be rare. In those cases, an ephemeral clone of the production environment should be created with infrastructure as code, and the dangerous changes or the intense testing can happen there.&lt;/p>&lt;p>Having dev, test, or staging environments is a red flag.&lt;/p>&lt;h2 id="sustainable-pace">Sustainable pace&lt;/h2>&lt;p>Working hours should include 5 hours a day of collaborative work, plus 2 hours of learning activities. Holidays should be unlimited, with a minimum of 21 days a year plus the national holidays (typically 9 days). Holidays booked past the 21 days minimum should not be recorded.&lt;/p>&lt;p>People should have flexibility in terms of working from home and working hours, but the expectation is that, on most days, during core hours, the whole team is synchronously working together in the same location.&lt;/p>&lt;h2 id="products-not-projects">Products, not projects&lt;/h2>&lt;p>The company should avoid projects, and work on products instead. The difference is that the work on products is continuous, without start and end dates, and with no specific customer in mind.&lt;/p>&lt;p>There&amp;rsquo;s also no need for deadlines, as the company should only sell what&amp;rsquo;s already built. Estimates become unnecessary as well, as the initiatives should be funded based on the value they&amp;rsquo;re producing and the pace they&amp;rsquo;re progressing at.&lt;/p>&lt;p>The company should set goals, expressed as qualitatively improved user behaviour, and the teams should produce ideas to reach these goals. The assumptions behind each idea should be validated, using the existing data, or through product discovery techniques. The most promising idea, based on the evidence, should be attempted first. This does not mean the whole idea should be built, only an initial step, to gain further validation that the idea will work out. If the initial step is working well in the hands of the customers, then a second small step should be attempted. Every task should be associated with a step.&lt;/p>&lt;p>This is, in a nutshell, the GIST framework codified by Itamar Gilad, and it&amp;rsquo;s an excellent way of running a product company. Mind that, in such an organization, funds get allocated fluidly based on demand, rather than being frozen upfront in large amounts. It&amp;rsquo;s the teams and the goals that get funded, not the ideas, the steps, or the tasks.&lt;/p>&lt;h2 id="employing-great-people">Employing great people&lt;/h2>&lt;p>The company should strive to ensure it employs great people. This should be reflected across many aspects, including salaries, benefits, but also on-the-job training. The acquired capabilities and the teams are the most important assets of a company, so they must be protected at all costs.&lt;/p>&lt;h2 id="roles">Roles&lt;/h2>&lt;p>Apart from product developers (of all sorts, backgrounds, calibre, specialty, etc.), the only other roles are:&lt;/p>&lt;ol>&lt;li>Cross-Functional Leads (CFLs): responsible to coordinate and shepherd a tribe. People with different skills, seniority, or expertise align to the cross-functional lead of the tribe they&amp;rsquo;re in. A cross-functional lead e.g., &amp;ldquo;Cross-Functional Lead, Savings&amp;rdquo;, can come from any background, even if experience in software development and product management are both heavily recommended. CFLs propose the product goals the company prioritizes.&lt;/li>&lt;li>Function Leads (FLs): responsible to coordinate work and initiatives within their function. People within that function liaise with the function lead, but are aligned to the cross-functional lead of the tribe they&amp;rsquo;re in. A function lead should be a very senior member of their function. As an example, the &amp;ldquo;Function Lead, Legal&amp;rdquo; should be a very senior corporate solicitor. Using the same example, legal experts in the various tribes would all align to their tribe&amp;rsquo;s CFL, but liaise with the Legal FL as well. FLs propose the technical goals the company prioritizes.&lt;/li>&lt;li>Managers: coordinators and shepherds of processes. Nobody manages people, only processes. Specifically, managers coordinate the continuous improvement of the processes they focus on, liaising with CFLs, FLs, and product developers. A manager should be intimate with the way the work their process aims at improving works. As an example, a software development manager should be intimate with how software development works, meaning that, in this example, that manager would have to have a strong background in software development.&lt;/li>&lt;li>Founders or CEO: responsible to protect the system of work, and to come up with and continuously repeat the vision and the mission, so that everybody aligns.&lt;/li>&lt;/ol>&lt;p>It&amp;rsquo;s important to stress again the fact that nobody is managed by anybody. If required, every employee formally reports to the CEO. CFLs, FLs, and managers coordinate the work of those teams closest to the customers, but this is nothing more than coordination. Managers are not subordinated to CFLs or FLs, it&amp;rsquo;s just a different focus. So thinking about CFLs and FLs as directors, under the executives, and above management is incorrect here.&lt;/p>&lt;h2 id="one-way-door-decisions-are-reviewed-and-can-be-vetoed">One-way door decisions are reviewed and can be vetoed&lt;/h2>&lt;p>Most decision can be easily reverted. These 2-way door decisions are made autonomously by the teams.&lt;/p>&lt;p>Some decisions, though, are more like one-way doors, meaning there&amp;rsquo;s no easy way to revert them, and their consequences can be significant. The teams must present their reasoning and conclusions to people in an inner circle, whenever they make a one-way door decision. The reviewers can veto the decision, explaining their reasoning, but cannot force a decision. The review board includes the relevant cross-functional lead and function leads.&lt;/p>&lt;p>If a team disagrees with the review board, they can escalate this further to the founders and/or the CEO.&lt;/p>&lt;p>In any case, whenever a decision is vetoed, the team must make a different decision, and repeat the process. CFLs, FLs, founders, and CEO shouldn&amp;rsquo;t make the decision themselves.&lt;/p>&lt;p>There are many kinds of one-way door decisions, and the teams need to be trained to identify them correctly. Firing people is a great example of something that is always a one-way door, and must thus be presented for review before a team can move forward.&lt;/p>&lt;h2 id="transparency-and-access-to-information">Transparency and access to information&lt;/h2>&lt;p>Most information should be accessible to everybody. This should include salaries, and even information employees could exploit to misbehave e.g., to commit insider trading. The damage lack of transparency causes is worse than any risks deriving from sharing dangerous information.&lt;/p>&lt;p>There should only be a few exceptions, including information that would paint a person under a bad light, and personal information e.g., health conditions, etc.&lt;/p>&lt;h2 id="vigorous-learning-on-the-job">Vigorous learning on the job&lt;/h2>&lt;p>Vigorous learning should happen as part of the job. Function-specific and cross-functional training should both be organized by the company for all employees. Training should be coordinated by both internal and external experts, and should include doing actual work outside of someone&amp;rsquo;s main function.&lt;/p>&lt;p>Training should involve theory and practice, and occur during working hours. Participation should be mandatory for everybody.&lt;/p>&lt;p>Cross-functional training for developers should include finance, marketing, product management, design, sales, business, and domain knowledge, e.g., industry-specific concepts. For other functions, it should be similar, and include software development.&lt;/p>&lt;p>20% of working hours is the minimum that&amp;rsquo;s effective, so that&amp;rsquo;s 1.5 hours a day or 1 day a week.&lt;/p>&lt;h2 id="under-performance-is-addressed">Under-performance is addressed&lt;/h2>&lt;p>A person or a team might sometimes under-perform. This can be noticed by anybody, including the person or team themselves. At first, the subject should try to diagnose and fix the problem on their own, asking anybody for help.&lt;/p>&lt;p>A variation of performance can be normal, or special. If it&amp;rsquo;s normal, then there&amp;rsquo;s no point in addressing anything, as things tend to fluctuate anyway. You can tell whether a measurement is within due to normal causes of variation by using control charts, but a subjective assessment is usually enough.&lt;/p>&lt;p>When the performance variation is deemed special, though, it remains to be seen whether it&amp;rsquo;s a temporary. If temporary, it might have to do with some external conditions: maybe a new joiner joined the team and the equilibrium will be restored soon, or a family situation is distressing a person, etc. The important thing in this case is identifying the probable causes, and monitoring how the situation evolves.&lt;/p>&lt;p>If it&amp;rsquo;s not a temporary condition, or no probable causes have been identified, it&amp;rsquo;s probably the result of the ways of working. Did any recent changes to the ways of working occur? Will the situation take care of itself? What&amp;rsquo;s working less effectively? Identifying the systemic pulls that might be causing the under-performance is important.&lt;/p>&lt;p>In case the person or team thinks they addressed the situation, they should ask the person who originally spotted the problem (could be themselves) whether they agree that the problem has been addressed effectively. If the answer is no, this should be escalated to management.&lt;/p>&lt;p>Whenever the person or team failed at addressing the situation on their own, managers should review what&amp;rsquo;s happening by joining the relevant team for a duration. The time this takes varies, but it&amp;rsquo;s rarely short. 2 weeks at a minimum for a person, and at least 1 month for a team. While embedded in the team, 2 managers will immerse themselves with everything the team does.&lt;/p>&lt;p>In a first phase, the managers will simply observe and take notes, with the occasional question. In a second phase, the managers might suggest improvements or other actions.&lt;/p>&lt;p>If no other action fixes the problem, the managers can decide to disband a team, merging each person into a different team. When it comes to people, skills gaps can be addressed most of the time, and alternative jobs can be tried within the same company. However, sometimes a person&amp;rsquo;s belief system makes them incompatible with the working principles described here, or even with software product development in general. In this case, the managers can decide to terminate that person&amp;rsquo;s employment.&lt;/p>&lt;p>In all cases, this is always a one-way decision, so it should be presented to the CFL for that area, and to the relevant FLs, as usual. If the CFL and the FLs veto the decision, as usual, the managers can decide to escalate to the founders and/or the CEO.&lt;/p>&lt;p>Whenever the person who&amp;rsquo;s under-performing is:&lt;/p>&lt;ol>&lt;li>A CFL or a FL, the procedure is identical, apart from the fact that the decision made by the 2 managers will be presented to the founders and/or CEO instead.&lt;/li>&lt;li>A manager, the procedure is identical, but different managers are involved.&lt;/li>&lt;li>A founder or CEO, the procedure is identical, apart from the fact that the decision made by the 2 managers will be presented to the board of the company. If there&amp;rsquo;s no board, and it&amp;rsquo;s a fully-owned private company, and the founder and CEO is under-performing, then the employees will have to live with that. That&amp;rsquo;s why boards are important.&lt;/li>&lt;/ol>&lt;p>If the person who reported the problem is:&lt;/p>&lt;ol>&lt;li>Part of the group the decision would be presented to, then this is escalated to an inner level instead.&lt;/li>&lt;li>A manager, the procedure is identical, but different managers are involved.&lt;/li>&lt;/ol>&lt;p>Whenever under-performance results in a person getting fired or a team getting disbanded, an email is sent to the whole company, explaining what happened, who made the decision, and what the reasoning was. This is to avoid rumors.&lt;/p>&lt;h2 id="bad-behaviour-is-addressed">Bad behaviour is addressed&lt;/h2>&lt;p>Bad behaviour differs from under-performance, so it&amp;rsquo;s important to address it differently. People or teams are bad actors when they damage others or the company with their actions, and they do so consciously.&lt;/p>&lt;p>The system of work, especially the incentives, greatly influences everybody&amp;rsquo;s behavior. And yet, when facing the same systemic pulls, some people decide to do the right thing, regardless the cost, while others easily give in and become bad actors. So, while under-performance is usually something fixable, and the company has a duty to try its best to address it, bad behavior is a different story.&lt;/p>&lt;p>It&amp;rsquo;s important to stress again that being aware of the damage they cause is a key requirement for somebody to qualify as a bad actor. Also, if somebody did something bad, then realized it, and reported it, they&amp;rsquo;re not bad actors. But if they did that while understanding the consequences, or if they decided to cover it up after realizing it, they need to be rooted out.&lt;/p>&lt;p>Bad actors are a fundamental threat to companies. While truly malicious actors are rare, selfish and toxic behaviors are quite common, can destroy morale and culture, and spread like mold when left unaddressed.&lt;/p>&lt;p>Anybody can report that they feel a person or a team behaved badly. Reporting should never happen anonymously, but management should keep the identity of the reporter a secret. This is an exception to the default transparency for all things, aimed at creating a safe space for these conversations. It&amp;rsquo;s also imperative that, whenever an escalation is deemed unfounded, the reporter isn&amp;rsquo;t negatively impacted. If the investigators suspect that the reporter acted maliciously in the first place, though, the reporter should also be investigated as a potential bad actor.&lt;/p>&lt;p>The investigation is conducted by 2 managers, who review the information reported, interview relevant people in private, collect the anecdotes and evidence in a file, and eventually speak to the person under investigation.&lt;/p>&lt;p>The person is informed of the investigation only if the initial anecdotes and evidence substantiate the claim. If that happens, before making a decision, the 2 managers speak with the subject of the investigation, in private. They explain what the claim is, without mentioning who the reporter is, explain how the process was run, and expose the anecdotes and evidence collected, striving not to reveal the names of the people interviewed. The person being investigated can then explain their reasoning, behavior, and version of the story. The 2 managers make a decision and, if the claim is deemed founded, proceed to seek validation from people at an inner level, as these are always one-way door decisions.&lt;/p>&lt;p>Whenever revealing anecdotes or evidence would betray their source, the managers should tell the person being investigated anyway, as their right to explain themselves takes precedence over the keeping all the people involved anonymous.&lt;/p>&lt;p>If the person under investigation is:&lt;/p>&lt;ol>&lt;li>A CFL or a FL, the procedure is identical, apart from the fact that the decision made by the 2 managers will be presented to the founders and/or CEO instead.&lt;/li>&lt;li>A manager, the procedure is identical, but different managers are involved.&lt;/li>&lt;li>A founder or CEO, the procedure is identical, apart from the fact that the decision made by the 2 managers will be presented to the board of the company. If there&amp;rsquo;s no board, and it&amp;rsquo;s a fully-owned private company, and the founder and CEO is a bad actor, then the employees will have to live with that. That&amp;rsquo;s why boards are important.&lt;/li>&lt;/ol>&lt;p>If the person who reported the problem is:&lt;/p>&lt;ol>&lt;li>Part of the group the decision would be presented to, then this is escalated to an inner level instead.&lt;/li>&lt;li>A manager, then the procedure is identical, but different managers are involved.&lt;/li>&lt;/ol>&lt;p>If the decision is deemed founded by management and validated by the review board, the person under investigation receives disciplinary action. In most cases, this should be termination of employment, as minor cases like a one-time hoarding of candies should be dismissed by the managers in the first place.&lt;/p>&lt;p>Whenever the investigation impacts a team, rather than a person, the disciplinary action should affect all the team members that were aware. The managers should investigate who was aware and include this as part of the decision that gets validated by the review board.&lt;/p>&lt;p>Everytime bad behavior results in a person getting fired or a team getting disbanded, an email should be sent to the whole company, explaining what happened, who made the decision, and what the reasoning was. This is to avoid rumors and reinforce the message that bad behavior is not tolerated.&lt;/p>&lt;h1 id="parting-words">Parting words&lt;/h1>&lt;p>In hindsight, this blog post turned out to be longer and much harder to write than I originally imagined.&lt;/p>&lt;p>What are your thoughts about these principles? Do you see eye to eye with me about them? Which ones give you an emotional response, either positive or negative? How many of these does your company adopt? Why?&lt;/p>&lt;p>As always, let me know your thoughts on LinkedIn or Medium!&lt;/p></description></item><item><title>At the heart of the burnout epidemic</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-11-06-at-the-heart-of-the-burnout-epidemic/</link><pubDate>Mon, 06 Nov 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-11-06-at-the-heart-of-the-burnout-epidemic/</guid><description>&lt;p>These days we&amp;rsquo;re dealing with a job burnout epidemic. Everybody speaks about it, there&amp;rsquo;s a lot of awareness, and yet things have been getting worse and worse every year. So what have we been missing?&lt;/p>&lt;p>Curious? I&amp;rsquo;m going to offer a contrarian take, as usual. So buckle up and keep reading.&lt;/p>&lt;h1 id="what-burnout-means">What burnout means&lt;/h1>&lt;p>Job burnout is a state of emotional, physical, or mental exhaustion, experienced because of work. Burnout can have serious consequences, including frequent illness, disengagement and detachment, insomnia, feelings of frustration, helplessness, anger or irritability, hopelessness, and loss of motivation.&lt;/p>&lt;p>How can you tell if you might be experiencing job burnout? Ask yourself:&lt;/p>&lt;ol>&lt;li>Have you become cynical at work?&lt;/li>&lt;li>Do you drag yourself to work and have trouble getting started?&lt;/li>&lt;li>Have you become irritable or impatient with co-workers, customers or clients?&lt;/li>&lt;li>Do you lack the energy to be consistently productive?&lt;/li>&lt;li>Do you find it hard to concentrate?&lt;/li>&lt;li>Do you dread Mondays?&lt;/li>&lt;li>Do you lack satisfaction from your achievements?&lt;/li>&lt;li>Do you feel disillusioned about your job?&lt;/li>&lt;li>Are you using food, drugs or alcohol to feel better or to simply not feel?&lt;/li>&lt;/ol>&lt;h1 id="the-usual-explanations">The usual explanations&lt;/h1>&lt;p>Common explanations for job burnout include:&lt;/p>&lt;ol>&lt;li>If you work long hours or have an excessive workload.&lt;/li>&lt;li>When a job is chaotic, you need constant energy to remain focused, which can lead to fatigue and job burnout.&lt;/li>&lt;li>Lack of rewards and acknowledgement. If you don&amp;rsquo;t get recognized and compensated enough for your efforts.&lt;/li>&lt;li>Poor self-care: not taking care of yourself outside of work.&lt;/li>&lt;li>Juggling many things at once: work, family, study, etc.&lt;/li>&lt;/ol>&lt;h1 id="everybody-talks-about-it">Everybody talks about it&lt;/h1>&lt;p>Perhaps unsurprisingly, job burnout is getting a lot of attention. Everybody talks about it.&lt;/p>&lt;p>Companies offer stress management courses, coaching, more holidays, etc. However, the reason why they care is purely pragmatic:&lt;/p>&lt;ol>&lt;li>Burned-out employees are 2.6 times more likely to be actively seeking a different job.&lt;/li>&lt;li>Burnout syndrome accounts for a large part of all occupational illness cases.&lt;/li>&lt;li>Burned-out workers are way more likely to take a sick day.&lt;/li>&lt;li>A report by Gallup and Workhuman discovered turnover and lost productivity due to employee burnout cost businesses around $322bn globally, with the cost of voluntary turnover due to burnout alone being 15% to 20% of the payroll budget each year.&lt;/li>&lt;/ol>&lt;h1 id="a-burnout-epidemic">A burnout epidemic&lt;/h1>&lt;p>And yet, job burnout is an epidemic. I don&amp;rsquo;t want to go into this at length, but here are a few bullet points that convey the pervasiveness of the problem:&lt;/p>&lt;ul>&lt;li>American Psychological Association: 3 in 5 workers reported experiencing negative mental and physical impacts due to work-related stress. 26% noted a lack of interest, motivation, or energy, 32% reported emotional exhaustion, and 44% noted high levels of physical fatigue.&lt;/li>&lt;li>Indeed: 52% of workers reported feeling burned out.&lt;/li>&lt;li>Deloitte: 77% of workers have experienced burnout at their current job.&lt;/li>&lt;li>Mental Health UK: 46% of workers felt &amp;ldquo;more prone to extreme levels of stress&amp;rdquo; compared to the year before, while a massive 1 in 5 reported feeling &amp;ldquo;unable to manage stress and pressure in the workplace&amp;rdquo;.&lt;/li>&lt;li>Mental Health America: 75% of employees experienced burnout.&lt;/li>&lt;/ul>&lt;p>And it&amp;rsquo;s getting worse, year-on-year burnout grows at a rate of 5% to 8%.&lt;/p>&lt;h1 id="the-popular-model-doesnt-make-sense">The popular model doesn&amp;rsquo;t make sense&lt;/h1>&lt;p>So why is burnout on the rise, despite the increasing efforts to address it? And despite the fact that working conditions are generally getting better?&lt;/p>&lt;p>Think about the factors that are commonly cited as the causes of burnout, and look at their trend over the past two decades:&lt;/p>&lt;ol>&lt;li>Working long hours and excessive workloads.&lt;/li>&lt;li>Unstructured work without clear goals.&lt;/li>&lt;li>Lack of rewards and acknowledgement.&lt;/li>&lt;li>Poor self-care.&lt;/li>&lt;li>Juggling many things at once.&lt;/li>&lt;/ol>&lt;p>Do you see what&amp;rsquo;s odd here? Haven&amp;rsquo;t all these things improved, compared to 20 years ago?&lt;/p>&lt;p>Aren&amp;rsquo;t we all working fewer hours per week? Hasn&amp;rsquo;t work become more structured? Aren&amp;rsquo;t companies more thoughtful about incentives, recognition, and bonuses? Or aren&amp;rsquo;t people getting more support with mental health days off, paid gyms, and private healthcare? And do we really think people have to juggle more than they used to juggle 20 years ago?&lt;/p>&lt;p>So why has burnout continued to rise, even when its stated causes have improved?&lt;/p>&lt;p>I believe the reason lies in a fundamental lack of understanding, resulting in an inability to see what the actual root causes of burnout are, and to effectively address the problem.&lt;/p>&lt;h1 id="companies-dont-get-it">Companies don&amp;rsquo;t get it&lt;/h1>&lt;p>Companies are worried about loss in productivity and the voluntary attrition. So they offer coaching, counseling, mental health days off, more holidays, more breaks, shorter weeks, more flexibility, helplines, and awareness programs.&lt;/p>&lt;p>But what do all these support mechanisms have in common? They are all directed at helping employees cope with the stress. It&amp;rsquo;s always &amp;ldquo;when you feel like this, this thing could help&amp;rdquo;.&lt;/p>&lt;p>Companies seem to believe that job burnout has to do with the employees, and how they poorly manage the stress they see as intrinsic to the work.&lt;/p>&lt;p>So it&amp;rsquo;s all about what the company should do about the employees. Never about what the company should do to eliminate the stress. Not about changing how the work works, so that stress isn&amp;rsquo;t an intrinsic condition, and so that it doesn&amp;rsquo;t eventually cause burnout in every employee.&lt;/p>&lt;h1 id="at-the-heart-of-the-problem">At the heart of the problem&lt;/h1>&lt;p>What if a state of continuous stress wasn&amp;rsquo;t intrinsic to work, and was somehow worse these days, compared to 20 years ago? At the heart of the problem, there must be a fundamental human need that is not met.&lt;/p>&lt;p>I believe that job burnout is the consequence of the need for self-actualization not getting met. People yearn to fully exercise their capabilities in pursuit of something meaningful.&lt;/p>&lt;p>Some folks find ways of meeting this need outside their main job. It&amp;rsquo;s somewhat acceptable, and these people are the few ones that tend to be protected from job burnout, to an extent. It&amp;rsquo;s still pretty miserable, as they cannot bring themselves to love their job, and only see it as a way of getting a paycheck. As a company with aspirations, you shouldn&amp;rsquo;t want these folks anyway.&lt;/p>&lt;p>For the majority, though, work is the only way of getting this fundamental need met. After all, most people work and commute for roughly two thirds of their waking hours.&lt;/p>&lt;h1 id="various-coping-mechanisms">Various coping mechanisms&lt;/h1>&lt;p>When this need is not met, people inevitably suffer and enter a state of perpetual stress. This can wreak havoc on a person&amp;rsquo;s mind, leading to various kinds of coping mechanisms.&lt;/p>&lt;p>Some folks feel under-challenged, not capable of fully leveraging their skills, of stretching, and of reaching further proficiency levels. They resent their job, and leave as soon as they find something they feel would be more challenging.&lt;/p>&lt;p>Others, having been robbed of their intrinsic motivation, pursuit the illusion of self-actualization by chasing recognition (bonuses, career advancement, etc.). They play the game, overwork, and do anything to advance. They hate their job as well, but compensate by getting recognised. They inevitably either run out of steam at some point, or leave the company for a promotion elsewhere. These people are also not good for a company.&lt;/p>&lt;p>And last, some other folks abandon all hopes that doing meaningful work is even possible. They get angry at things and at people, hate their job, their company, and their co-workers, and stop caring about anything at all. This last type of burned-out people tend to stay in the company, as they&amp;rsquo;re disillusioned enough not to believe it could be better anywhere else. As a company with aspirations, these are arguably the people you should want the least.&lt;/p>&lt;h1 id="whats-preventing-self-actualization-at-work">What&amp;rsquo;s preventing self-actualization at work&lt;/h1>&lt;p>There are three components that must co-exist, for the self-actualization need to be met:&lt;/p>&lt;ol>&lt;li>The work must be in pursuit of something meaningful.&lt;/li>&lt;li>The work must exercise the person fully, stretch their limits and, as a consequence, develop them.&lt;/li>&lt;li>The person must feel their work contributes to advancing the company towards the meaningful objective.&lt;/li>&lt;/ol>&lt;p>They are all important, and all are suffering more and more these days. Finding a job where these aspects co-exist is becoming increasingly harder, across most professions, and certainly in software product development.&lt;/p>&lt;p>About the meaning of work, it&amp;rsquo;s rarer and rarer for companies to aim at greatness, to tackle important problems, and to advance society and the human condition.&lt;/p>&lt;p>The division of labor is increasingly restricting the boundaries of employees&amp;rsquo; contribution. Companies are more and more divided into people who think and decide, people who operate and manage, and people who execute without questions. There&amp;rsquo;s also a widespread intentional push for mediocrity and good enough; the days when excellence was encouraged or even expected are long gone.&lt;/p>&lt;p>Teamwork has lost its meaning. Nowadays, it does not mean working together with others towards a shared goal. It&amp;rsquo;s about working individually, exclusively within a domain area you happen to share with other individual contributors.&lt;/p>&lt;p>Overall, people find themselves doing work that doesn&amp;rsquo;t challenge or excite them, with people they don&amp;rsquo;t trust, following plans and decisions made by managers and executives they don&amp;rsquo;t respect.&lt;/p>&lt;h1 id="parting-words">Parting words&lt;/h1>&lt;p>Job burnout is an epidemic, and it&amp;rsquo;s getting worse. Companies believe that stress is intrinsic to work, so they offer coping mechanisms and support. Instead, it&amp;rsquo;s how the work works these days that creates this perpetual condition of stress, by not meeting the self-actualization human need.&lt;/p>&lt;p>The impact this has is tremendous. Companies superficially worry about productivity, or how quickly and efficiently people do what they&amp;rsquo;re told. They don&amp;rsquo;t realize that burnout damages contribution the most. Burned-out people don&amp;rsquo;t actively shape the work and how the work works, so companies cannot tap into the collective skills and intelligence of their employees.&lt;/p>&lt;p>After all you read, do you still think job burnout is normal? Or do you think now believe it&amp;rsquo;s the result of ineffective and anti-systemic ways of working? If you&amp;rsquo;re experiencing burnout, or if you see burned-out people at your company, do you now see what the problem is?&lt;/p></description></item><item><title>The product manager role is a mistake</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-21-product-manager-role-is-a-mistake/</link><pubDate>Sat, 21 Oct 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-21-product-manager-role-is-a-mistake/</guid><description>&lt;h1 id="the-popular-model">The popular model&lt;/h1>&lt;p>Software product companies aren&amp;rsquo;t usually started by people with a strong background in building software products. Those who most likely end up founding businesses are good at getting funding, and at understanding their target industry. They&amp;rsquo;ve been in their industry for a long time, know their industry inside out, convinced themselves this means they can do it, and used their relationships and skills to get the money necessary to start a company.These people tend to lead everything personally, and a lot could be said about the damage domain experts cause to their own companies when they lead things. Somebody already did it though, and they did it so well that there&amp;rsquo;s no point in me doing a worse job.&lt;/p>&lt;p>Sometimes, though, they acknowledge the danger they represent, and hire somebody in charge of making decisions. Often, they think &amp;ldquo;We don&amp;rsquo;t want to be a service organization, but a product company. Product companies are product-led, hence product managers should be in charge.&amp;rdquo; (This is a terrible misunderstanding, but a popular one: product-led does not mean having PMs in charge.)&lt;/p>&lt;p>So they go and hire a career product manager, and have them make all decisions. Whenever somebody objects to this idea, the usual answer is that &amp;ldquo;all great companies work like this: Amazon, Apple, Tesla, etc.&amp;rdquo;.&lt;/p>&lt;p>It&amp;rsquo;s unquestionable that many great companies delegate control to a single person. Indeed, not all great companies adopt this model, but let&amp;rsquo;s keep this aspect for the end of the article.&lt;/p>&lt;h2 id="tremendous-people-are-amazing">Tremendous people are amazing&lt;/h2>&lt;p>So what&amp;rsquo;s wrong with giving control of a company to a single person? The problem is not much with the model, which can work really well, but with the who, the person you give control to.&lt;/p>&lt;p>Humans are inevitably fascinated by greatness. Movies, books, documentaries, quotes: we all can&amp;rsquo;t help being inspired by those people that changed the world with their vision and work.&lt;/p>&lt;p>Picasso, Jeff Bezos, Mozart, Steve Jobs, Nikola Tesla, Elon Musk, Robert Oppenheimer, Peter Thiel, Leonardo, and others are worshipped by the masses. And what&amp;rsquo;s not to love in the inspiring vision, magnetic charisma, unwavering self-confidence, stalwart resilience, and sheer greatness these people embody? Many would do anything to work with these people. Heck, for these people.&lt;/p>&lt;p>Tremendous people envision alternative realities, fully commit themselves to their work, and move mountains with their willpower. Even if these figures are often controversial, everybody dreams of being like them. You can love them or hate them, but you&amp;rsquo;re forced to acknowledge them.&lt;/p>&lt;p>So what&amp;rsquo;s the problem with wanting to adopt this model? Absolutely nothing. It&amp;rsquo;s amazing.&lt;/p>&lt;h2 id="but-theyre-not-for-every-company">But they&amp;rsquo;re not for every company&lt;/h2>&lt;p>Do you know, though, what all the great people mentioned above have in common? None of them was a product manager. Not a single one. Actually, take a minute to think hard, and come up with one product manager that achieved anything comparable to these folks.&lt;/p>&lt;p>These people were artists, mathematicians, inventors, scientists, CEOs, and often more than one of these at a time. Do you know what else they had in common? None of them were motivated by OKRs, KPIs, data, or other people&amp;rsquo;s opinion. They were obsessed by their vision and craft. In an all-consuming, totally uncompromising way: anything short of perfection was inconceivable to them.&lt;/p>&lt;p>Tremendous people also tend to have very strong opinions, a big ego, their way of doing things, rough edges, and a disdain for mediocrity. They&amp;rsquo;re borderline unemployable, and they likely wouldn&amp;rsquo;t want to work in most companies. To them, it&amp;rsquo;s about creating something extraordinary for the sheer pleasure of it. And that&amp;rsquo;s why they disregarded work-life balance: their craft is their biggest joy. Imagine asking Michelangelo not to sculpt the back of a statue because the customer wants to put it against a wall.&lt;/p>&lt;p>But the more somebody is mediocre, the more likely they are to be put off by this intensity and passion. The more they&amp;rsquo;re willing to compromise, to lower standards, and to cut corners.&lt;/p>&lt;p>There are various degrees of greatness, but the greater somebody is, the more they approach the behavior of the greatest.&lt;/p>&lt;p>So, sure, if you can find someone tremendous, you&amp;rsquo;re willing to get out of their way, and you can cope with their intensity, by all means give them control over your product and company. They&amp;rsquo;ll produce extraordinary things.&lt;/p>&lt;p>Good chances are, though, that you wouldn&amp;rsquo;t be able to find them, that they wouldn&amp;rsquo;t want to work for you or with you, and that you wouldn&amp;rsquo;t be able to tolerate their ego and burning intensity.&lt;/p>&lt;h2 id="the-mistake">The mistake&lt;/h2>&lt;p>So what do you do instead? The enormous mistake you can make is to adopt a model that requires tremendous people, without having these tremendous people. The product managers you can hire and are comfortable hiring aren&amp;rsquo;t tremendous people. Whether they&amp;rsquo;re business lackeys, framework fanatics, subject-matter experts, former makers that climbed the corporate ladder, metrics junkies, or - God forbid - MBA graduates, there&amp;rsquo;s nothing great about them.&lt;/p>&lt;p>So what happens when you try to replicate something you have heard or read, but with a different breed of people? When you think &amp;ldquo;product led&amp;rdquo; means having a product manager making all decisions, and you hire a career product manager for this?&lt;/p>&lt;p>It can be deadly, and it typically is. And not the kind of death people can usually see coming. A much sneakier one, one that saps your company&amp;rsquo;s strength and empties its soul, until there&amp;rsquo;s nothing left.&lt;/p>&lt;p>These PMs you hire have no vision, charisma, expertise, strong beliefs, or technical skills. So what do they do when you give them that kind of power? They play it safe, and they play politics. Since they know they&amp;rsquo;re inadequate, they continuously attempt to claim credit and avoid blame.&lt;/p>&lt;p>First, they weaponize data-driven decisions. Instead of acknowledging that you can find supporting data for every claim, they pretend that their decisions are the only possible sensible possibilities given the data. When things go south, they blame the market, the quality of the data, or the company&amp;rsquo;s bad luck. It was never their fault, as that&amp;rsquo;s what the data was saying.&lt;/p>&lt;p>Next up is introducing frameworks and roles for everything. You get a career progression framework with 18 levels, quarterly performance assessments, PI planning, ADRs, PRDs, RACI, MOSCOW, OKRs, KPIs, DORA, and SAFe. People get blamed for not following the process, not being data-driven, not writing enough documentation, not keeping their documentation up-to-date, and so on. You also end up with Director of KPIs, Deputy Director of Product Operations, and Global Head of Digital Whiteboard Solutions. You soon need approvals for everything, and nobody entirely controls a process or aspect anymore. Everything involves a large committee, where anyone can say &amp;ldquo;no&amp;rdquo; to anything, but nobody can fully say &amp;ldquo;yes&amp;rdquo;.&lt;/p>&lt;p>Meetings also tend to explode in number: status reports, alignment talks, 1:1s, staff meetings, strategy discussions, context-setting talks, release go/no-go meetings, project reports, etc.&lt;/p>&lt;p>As a final nail in the coffin, they hire a ton of bad people. People are hired based on whether they&amp;rsquo;d represent no threat, so any trace of greatness pretty much guarantees a rejection. The ideal hire becomes a mindless all-rounder, who semi-competently does everything they&amp;rsquo;re told, without any dissent or care for quality. They need hordes of them, to hide among the pack, and so that there are always people to blame. The number of product managers, in particular, skyrockets. A whole hierarchy of PMs, one or more in each team and on each aspect, all behaving the same.&lt;/p>&lt;h2 id="the-damage">The damage&lt;/h2>&lt;p>What happens to companies that, by putting mediocre people in charge, start the journey described above? Well, it&amp;rsquo;s not pretty.&lt;/p>&lt;p>Anyone with even a speckle of greatness eventually leaves. They either get pushed out by the mediocre dictator, or they leave on their own terms, as the environment has become intolerable to them.&lt;/p>&lt;p>Work grinds to a halt. Doing anything is suddenly painful, a big commitment, and a risky endeavor. Everybody runs just to stand still. Any grand vision becomes so unrealistic that it&amp;rsquo;s quickly abandoned. Low-hanging fruits and safe bets become the only work the company does. Even those who are not passionate about their craft feel the pain and eventually leave. Those who are numb enough to remain make the problem worse.&lt;/p>&lt;p>Slowly but surely, the company loses its best talent, forgets its vision, becomes internally-focused, and drifts into irrelevance. At best, it dies at some point. At worst, it survives in this half-life, turning passionate people into zombies who hate their job.&lt;/p>&lt;h1 id="a-ray-of-hope">A ray of hope&lt;/h1>&lt;p>Is there any alternative, then, to finding, hiring, putting in charge, and coping with a tremendous person? If you&amp;rsquo;re not pushed by greatness and craftsmanship, can you not build a successful product company? Good news is, you can! There&amp;rsquo;s an alternative model that works.&lt;/p>&lt;p>You hire everyday great people. Each one needs to be great at something, obsessed by their craft, and driven by quality. You then put them together in a team, without individual responsibilities, ensuring that there&amp;rsquo;s minimal overlap in areas of greatness. And you pay them to have fun doing what they love doing, in exchange for putting their time and skills at the service of your company.&lt;/p>&lt;p>These people are still going to be challenging, stubborn, egotistic, quirky, and volatile. But give them good problems to solve, allow them to do great work, and they&amp;rsquo;ll fully dedicate themselves to making your company successful.&lt;/p>&lt;h2 id="what-about-leaders-and-product-managers">What about leaders and product managers?&lt;/h2>&lt;p>Do you need to give up the idea of leaders altogether? Not really, even if you can definitely live without. You could avoid team leads completely, and have area leads as outside-in coordinators. But if you decide that you prefer having a leader responsible for each area, then by all means put a great person in that role. The leader of an area should be the person that gets closer to the various Jobs, Musk, and Bezos, in terms of vision, charisma, intensity, passion, and extreme behavior. Don&amp;rsquo;t fall into the trap of putting in charge those people who make your life easy in the short term, do what they&amp;rsquo;re told, and don&amp;rsquo;t require your attention. A wrong move here, and you risk the situation above.&lt;/p>&lt;p>And what about product managers? Is there no space for PMs in a well-operating company? Not for the PM role, no. At least not in its traditional meaning. This model still requires great people, passionate and visionary about their area of expertise. And you do need people that obsess about product discovery, product observability, and other areas traditionally delegated to product managers. Almost inevitably, these people will have held a product manager role, and that&amp;rsquo;s fine. But you&amp;rsquo;re hiring because of the obsession and the strengths, not because of the title. And you&amp;rsquo;re not going to give these folks a PM role, but put them in a team with people who obsess over engineering, design, performance, business, and domain knowledge.&lt;/p>&lt;h1 id="how-much-of-this-was-specific-to-product-managers">How much of this was specific to product managers?&lt;/h1>&lt;p>I&amp;rsquo;d be dishonest if I wouldn&amp;rsquo;t admit that the problems I described happen everytime anyone without vision, charisma, passion, technical skills, and willpower is put in charge. So why the focus on product managers? It&amp;rsquo;s because of this tendency to wanting people with a product manager title in charge of software product development. To be fair, some PMs I know are absolutely tremendous, and would do very well in both the models I described. But they wouldn&amp;rsquo;t necessarily do better than other great people, regardless of their background. Also, they&amp;rsquo;re a tiny minority, and most career PMs are nowhere near being capable of leading by greatness.&lt;/p></description></item><item><title>You should fear competition more than you fear your competitors</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-19-on-competition/</link><pubDate>Thu, 19 Oct 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-19-on-competition/</guid><description>&lt;h2 id="the-situation">The situation&lt;/h2>&lt;h3 id="an-obsession-with-competitors">An obsession with competitors&lt;/h3>&lt;p>Most startups obsess over their competitors. They watch every move, rush to copy every initiative or feature, analyze their marketing, check whether they&amp;rsquo;re being outsold, and keep an eye on others&amp;rsquo; funding rounds.&lt;/p>&lt;p>But is it useful? I believe competitors are a largely overrated threat, and that the obsession with competition is a much more dangerous killer.&lt;/p>&lt;p>In his book &amp;ldquo;The Mom Test: How to talk to customers &amp;amp; learn if your business is a good idea when everyone is lying to you&amp;rdquo;, Rob Fitzpatrick says: &amp;ldquo;They say that startups don’t starve, they drown. You never have too few options, too few leads, or too few ideas. You have too many. You get overwhelmed. You do a little bit of everything.&amp;rdquo;&lt;/p>&lt;p>If you think about it, software product development is not a race with other companies. Of course, sometimes there are new untapped opportunities where first-mover advantage is big, but most of the time it isn&amp;rsquo;t the case.&lt;/p>&lt;p>Good chances are, if what you&amp;rsquo;re doing is important, and nobody has done it before, no other company is going to jump on it just because they hear your idea. Also, two companies can never be the same: the internal dynamics, the people, the culture, the resources: all different. So it&amp;rsquo;s unlikely they&amp;rsquo;re going to produce comparable products.&lt;/p>&lt;h3 id="a-single-player-game-you-can-still-lose">A single-player game you can still lose&lt;/h3>&lt;p>If you think about it, software product development is a single player game. If you win big, but some other company wins bigger, nothing bad happens. There are no negative consequences in being the number 2 or even the number 30 in your industry. Breaking even is enough to guarantee a startup&amp;rsquo;s survival. For large companies, the threat of hostile takeovers or negative perception on the market is there, but if you&amp;rsquo;re not a public company, any profitability is enough.&lt;/p>&lt;p>Even if software product development is a single player game, it doesn&amp;rsquo;t mean it&amp;rsquo;s easy. You can still lose. In fact, you can beat all your competitors and still fail to capture value from your chosen market in exchange for the value you deliver.&lt;/p>&lt;p>Yes, because delivering value to your market in exchange for the resources (money, trust, etc.) you need to survive is what the game is about. Software product development is essentially a survival game.&lt;/p>&lt;h3 id="focusing-on-the-wrong-things">Focusing on the wrong things&lt;/h3>&lt;p>And yet, most startups focus on the wrong things: outputs, growth, investment, metrics, efficiency, revenue, closing deals, market size, and their competitors.&lt;/p>&lt;p>Don&amp;rsquo;t get me wrong, there&amp;rsquo;s nothing wrong with caring about these aspects. Actually, it&amp;rsquo;s necessary. The problem arises when other aspects, which are more important, get ignored or neglected, in pursuit of these.&lt;/p>&lt;p>Yes, because for startups and SMBs, the obsession should be about outcomes, development, sustainability, culture, people, effectiveness, profits, market characteristics, and delivering value.&lt;/p>&lt;h2 id="the-why">The why&lt;/h2>&lt;p>Let me be clear here: overlooking these aspects to focus on the former is a deadly mistake. I&amp;rsquo;d say it&amp;rsquo;s the fundamental reason why so many startups die within their first five years.&lt;/p>&lt;p>Why do so many companies obsess over the wrong things then? This is so ubiquitous that there must be systemic attractors at play. There have to be some conditions intrinsic to human nature or society which heavily skew the probability in favor of this dynamic.&lt;/p>&lt;p>And, as it turns out, there are.&lt;/p>&lt;p>&lt;!-- raw HTML omitted -->WARNING: what follows is a functional generalization, a conceptual model. Models are good when they&amp;rsquo;re useful, and to be useful they don&amp;rsquo;t need to be true.&lt;!-- raw HTML omitted -->&lt;/p>&lt;h3 id="salespeople-and-creators">Salespeople and creators&lt;/h3>&lt;p>At their core, you could characterize every person as either a salesperson or a creator. It&amp;rsquo;s not a binary thing, but most people tend to be heavily skewed in one direction.&lt;/p>&lt;h4 id="salespeople">Salespeople&lt;/h4>&lt;p>Salespeople worship competition. Their entire view of the world revolves around it.&lt;/p>&lt;p>The salespeople you encounter in SaaS companies tend to have been athletes, to have graduated from business school, and to have experienced service organizations. Their self-esteem comes from demonstrably being better than others. No wonder salespeople are naturally obsessed with quantifiable things and metrics. It&amp;rsquo;s all about KPIs, OKRs, P&amp;amp;L, and individual contribution.&lt;/p>&lt;p>So, naturally, they can&amp;rsquo;t help obsessing over their competitors, whether real or imaginary. There has to be an enemy to outsell. If anyone does better, it means they&amp;rsquo;re not the best, and being the best is what matters.&lt;/p>&lt;h4 id="creators">Creators&lt;/h4>&lt;p>Creators, on the other hand, cherish getting better at what they do. Their world revolves around perfecting their craft.&lt;/p>&lt;p>The creators that end up in SaaS businesses would have been geeks at school, graduated in a STEM subject, and experienced product organizations. What gets creators out of bed in the morning is practicing their craft to the limit of their skills with people they respect.&lt;/p>&lt;p>And it can only follow that creators obsess over their creations, rather than focusing on competition. The enemies are laziness, entropy, sloppiness, and stupidity. If their product isn&amp;rsquo;t well-crafted or useful, it means they&amp;rsquo;re not great at their craft, and that&amp;rsquo;s what matters to them.&lt;/p>&lt;h4 id="they-dont-like-each-other-much">They don&amp;rsquo;t like each other much&lt;/h4>&lt;p>Creators and salespeople have a hard time respecting each other.&lt;/p>&lt;p>Creators see salespeople as empty and soulless, insensible to quality, and incapable of producing anything useful. Creators tend to see marketing and sales as a necessary evil, fundamentally lies. In their ideal world, people would simply buy their creations based on their sheer quality. Creators often get pushed by salespeople to take shortcuts, feeling like Michelangelo would have felt if he was asked not to work on the back of his statues, as they&amp;rsquo;d be placed against a wall after all.&lt;/p>&lt;p>Salespeople can&amp;rsquo;t help seeing creators as naive. If only these geeks would be more pragmatic, they could close more deals. Customers demand features, so geeks should build these features, build them fast, and that&amp;rsquo;s all there is to it.&lt;/p>&lt;h4 id="every-company-needs-both">Every company needs both&lt;/h4>&lt;p>Regardless of their mutual disrespect, every company needs both salespeople and creators. Ironically, both are right. Salespeople are empty, and creators are naive. Or better, most people tend to strike a balance eventually, while a company too skewed in one direction might develop these traits.&lt;/p>&lt;p>While you need both creators and salespeople, what&amp;rsquo;s key is placing them in the right roles. Creators best cover management roles, as the outcome is the management system, how the work works. Salespeople, on the other hand, tend to be great in customer-facing roles, where the game is building relationships.&lt;/p>&lt;p>Slightly off-topic, but should managers for customer-facing roles be salespeople or creators? They should be creators really, otherwise you end up with managers thinking that their job is to be the best salesperson in the department.&lt;/p>&lt;h4 id="as-founders">As founders&lt;/h4>&lt;p>Walking up the management chain, you encounter founders and the CEO at the top. Now, founders are typically from different backgrounds, and this is a wonderful thing. You need diversity of strengths after all.&lt;/p>&lt;p>What about the CEO though? In a perfect world, their background wouldn&amp;rsquo;t matter much, as the chief executive officer isn&amp;rsquo;t the only executive officer. In reality, though, founder CEOs tend to see themselves as the owner of the company, managing the other executive officers. And both creators and salespeople push their characteristics to the extreme when in the top job. So whether a founder CEO is a salesperson or a creator tends to matter enormously.&lt;/p>&lt;h5 id="salespeople-1">Salespeople&lt;/h5>&lt;p>Salespeople, as CEOs, think their strength comes from their domain and business knowledge. They might have been in their industry for a long time, and know it inside out. They know all the big players, whether they&amp;rsquo;re customers, suppliers, or competitors.&lt;/p>&lt;p>Their obsession with metrics often materializes as wanting to manage the various areas of the company by objectives. Everything becomes a return on investment discussion. What&amp;rsquo;s not measurable inevitably suffers.&lt;/p>&lt;p>Most salespeople become founder CEOs in pursuit of money. The company is a means to an end. The ultimate goal is always proving to themselves and others that they&amp;rsquo;re the best, and making a lot of money is the ultimate proof.&lt;/p>&lt;p>So when salespeople create a company, they often have the exit already in mind. They raise investment from VCs, obsess over growth, and aim at getting bought or going public. Even when they want to keep owning a business, the aim is maximizing the cash extracted as dividends to the shareholders.&lt;/p>&lt;h5 id="creators-1">Creators&lt;/h5>&lt;p>When creators become CEOs and founders, they rarely possess deep domain knowledge in their chosen industry. What they tend to know is how to build software, and how products are discovered. The obsession is with craftsmanship and design.&lt;/p>&lt;p>Most creators become CEOs because they are disenchanted with the industry, and want to do their own thing to do it right. The company is a means to an end, and the goal is to practice their craft with people they respect. Making money is a necessary condition to continue exercising their craft.&lt;/p>&lt;p>So when creators found a company, they&amp;rsquo;re in for the journey. They hopefully get profitable sooner rather than later, and then just focus on the product. Revenues and growth are not the goal, and some companies purposely avoid growing their business beyond a certain limit, so they can keep enjoying what they do and how they do it.&lt;/p>&lt;h2 id="the-trillion-dollar-question">The trillion-dollar question&lt;/h2>&lt;p>The trillion-dollar question is now whether a SaaS software product company would be better off with a salesperson or a creator as founder and CEO. I say trillion-dollar question to indicate that it&amp;rsquo;s important but, if you consider that only 20% of SaaS companies survive their first 5 years in business, and only 8% survive their first 10 years, it&amp;rsquo;s easy to understand that this is an understatement.&lt;/p>&lt;p>Salespeople tend to get the top job because of how raising capital and acquiring customers work, and you tend to need money to start a business.&lt;/p>&lt;p>However, a salesperson CEO will almost always end up producing a service organization, as that&amp;rsquo;s all they have experienced and understand. This inevitably eventually fails, as service organizations are not structured to produce successful SaaS products, and they can&amp;rsquo;t generate the ROI investors expect for the money they put in the business.&lt;/p>&lt;p>Instead of me explaining why this is the case, I invite you to read the &lt;a href="https://bootcamp.uxdesign.cc/fail-to-scale-part-1-ad136e94b993">astonishingly amazing 4-part post by Jezz Santos&lt;/a>, especially &lt;a href="https://bootcamp.uxdesign.cc/fail-to-scale-part-3-468670356b88">the third part&lt;/a>.&lt;/p></description></item><item><title>Can accountability exist without authority?</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-15-responsibility-and-authority/</link><pubDate>Sun, 15 Oct 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-15-responsibility-and-authority/</guid><description>&lt;p>The ubiquitous way of structuring companies is by nesting areas of responsibility. You introduce well-defined areas and problems, and assign them to people, so that each area or concern has exactly one person responsible for it.&lt;/p>&lt;p>The nesting bit is because these areas roll up into wider areas. If someone is responsible for hiring, they might decide to further spread this responsibility across geographies. So you might have someone responsible for hiring software developers in London, rolling up into a UK hiring lead, rolling up into a global talent partner, rolling up into a chief operating officer.&lt;/p>&lt;p>I&amp;rsquo;m not a fan of this model, but it wouldn&amp;rsquo;t be that bad if it stopped at responsibility. The real problem arises when people are made accountable for the area they get assigned to.&lt;/p>&lt;p>Being accountable means that if the outcomes in your area are unsatisfactory, you get blamed for the situation. What&amp;rsquo;s the problem with accountability? Isn&amp;rsquo;t having people accountable kind of natural if we divide the areas of responsibility?&lt;/p>&lt;h2 id="the-problem-with-accountability">The problem with accountability&lt;/h2>&lt;p>There are several issues with dividing the responsibility across areas and then having the people responsible accountable for the results.&lt;/p>&lt;h3 id="interdependencies-between-the-areas">Interdependencies between the areas&lt;/h3>&lt;p>First of all, dividing the whole into parts, and assigning authority over each part to a person doesn&amp;rsquo;t work, because the parts are interdependent. In order to get good outcomes overall, which is the ultimate goal, the various parts must act in concert.&lt;/p>&lt;p>This is why in no company ever you&amp;rsquo;re fully in charge of an area, and you can make all decisions. It&amp;rsquo;s because no area is independent.&lt;/p>&lt;p>If you work in sales, and the product you sell is not great, your sales will be impacted. If your sales department struggles, and revenues are not what they should be, your budget in other areas is going to constrain what&amp;rsquo;s possible.&lt;/p>&lt;h3 id="accountability-cannot-exist-without-authority">Accountability cannot exist without authority&lt;/h3>&lt;p>Secondly, accountability cannot exist without authority. If I&amp;rsquo;m accountable for the outcomes, I cannot be merely an advisor. I must be the decision maker.&lt;/p>&lt;p>If I tell you what we should do, but when we disagree we do it your way, then I&amp;rsquo;m an advisor, and you should be accountable for the outcomes.&lt;/p>&lt;p>In almost all companies, authority follows the same hierarchy that responsibility gets assigned according to. So if you&amp;rsquo;re responsible for an area, your manager, who&amp;rsquo;s responsible for the containing area, has authority over you and your area. All the way up to the CEO.&lt;/p>&lt;p>And nobody can be held accountable for the outcomes if they have no authority.&lt;/p>&lt;h3 id="discontinuous-authority-doesnt-work-either">Discontinuous authority doesn&amp;rsquo;t work either&lt;/h3>&lt;p>For someone to be held accountable, there can be no exceptions to their authority. People who used to be accountable can become advisors, but once overruled, one cannot be held accountable again.&lt;/p>&lt;p>So if there&amp;rsquo;s a left or right turn, I advise to go right, and you choose to go left, you become accountable from that moment onward. You cannot abdicate right after, leaving the helm in my hands, and saying that I&amp;rsquo;m accountable.&lt;/p>&lt;p>Because there might not be a viable outcome down the path we just took. And going back to turn right might even be impossible. So I can become responsible for the area, focusing all my efforts there, but I cannot be accountable for the results, because by making that choice you constrained what results are possible.&lt;/p>&lt;p>Constraints of various kind define the situation and limit what&amp;rsquo;s achievable.&lt;/p>&lt;p>Imagine you outsourced product development for 4 years, built a low-quality product and hard-to-maintain systems, raised too much money or too little, too early or too late, and you&amp;rsquo;re now facing the consequences of your actions. It&amp;rsquo;d be disingenuous to put somebody in charge of product development and expect them to turn the situation on its head, without being ready for drastic measures.&lt;/p>&lt;h2 id="the-alternative">The alternative&lt;/h2>&lt;p>People can be asked to direct various areas, and given authority over them. But since this authority is always limited, various constraints limit what&amp;rsquo;s possible, and the outcomes are the product of the interactions of the various areas, holding somebody accountable for things they cannot control is naive and immoral.&lt;/p>&lt;p>The alternative to this anti-systemic approach is recognizing that in every company the CEO is accountable for everything. The remaining directors should be co-responsible for the direction of the company, and each responsible for their own area. In terms of efforts, not in terms of accountability over the outcomes.&lt;/p></description></item><item><title>The 3M Rule of management</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-07-on-managing-and-measuring/</link><pubDate>Sat, 07 Oct 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-10-07-on-managing-and-measuring/</guid><description>&lt;p>Metrics and indicators are amazing when used appropriately. But problems arise when people either:&lt;/p>&lt;ol>&lt;li>Use metrics as targets. I wrote a &lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-09-management-by-objectives-is-wishful-thinking/">post about this&lt;/a>, so I&amp;rsquo;m not going to explain why this is the case again.&lt;/li>&lt;li>Focus exclusively on things that can be measured, while neglecting things that cannot.&lt;/li>&lt;li>Adopt metrics without understanding their context. I wrote a &lt;a href="https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-31-developer-productivity-fuss/">post about McKinsey&amp;rsquo;s software development productivity nonsense&lt;/a>, so I&amp;rsquo;m not going to expand on this either.&lt;/li>&lt;li>Don&amp;rsquo;t factor in any delays between inputs and outputs.&lt;/li>&lt;li>Believe they need metrics in order to manage anything.&lt;/li>&lt;/ol>&lt;p>Today I want to challenge the last point, that you need metrics in order to manage anything.&lt;/p>&lt;h2 id="the-trend">The trend&lt;/h2>&lt;p>Companies are developing an obsession with metrics. &amp;ldquo;OKRs&amp;rdquo;, &amp;ldquo;KPIs&amp;rdquo;, and &amp;ldquo;data-driven&amp;rdquo; are on everyone&amp;rsquo;s mouth. Most companies even go as far as creating roles, teams, and departments solely responsible for defining and tracking metrics across the whole company.&lt;/p>&lt;p>This inevitably introduces problems, and when people bring these up and complain, they&amp;rsquo;re usually met with the weaponized quote that &amp;ldquo;you cannot manage what you cannot measure&amp;rdquo;.&lt;/p>&lt;h2 id="the-3m-rule-of-management">The 3M Rule of Management&lt;/h2>&lt;p>On the contrary, I propose the following 3M Rule of Management:&lt;/p>&lt;blockquote>&lt;p>&amp;ldquo;If you cannot Manage what you cannot Measure, then you&amp;rsquo;re a Muppet.&amp;rdquo;&lt;/p>&lt;/blockquote>&lt;h2 id="a-bit-of-theory">A bit of theory&lt;/h2>&lt;p>Before we dig deeper into why I believe you shouldn&amp;rsquo;t need metrics to manage, let&amp;rsquo;s go over a few definitions.&lt;/p>&lt;h3 id="measurement">Measurement&lt;/h3>&lt;blockquote>&lt;p>The process of associating numbers with physical quantities and phenomena.&lt;/p>&lt;/blockquote>&lt;p>Measurements concern themselves with quantities. Not every type of feedback is quantitative. Keep this in mind, because it&amp;rsquo;ll be important in a bit.&lt;/p>&lt;h3 id="management">Management&lt;/h3>&lt;blockquote>&lt;p>The process of dealing with or controlling processes, things, or people.&lt;/p>&lt;/blockquote>&lt;p>I&amp;rsquo;d argue that management is exclusively about processes, but let&amp;rsquo;s not diverge. Management is about steering something to produce desired outcomes.&lt;/p>&lt;h3 id="muppet">Muppet&lt;/h3>&lt;blockquote>&lt;p>Silly or unwise; showing poor judgment or little intelligence.&lt;/p>&lt;/blockquote>&lt;p>With the word &amp;ldquo;muppet&amp;rdquo;, I mean someone that struggles with boilerplate and routine activities.&lt;/p>&lt;h2 id="a-counter-example">A counter-example&lt;/h2>&lt;p>If it was true that &amp;ldquo;you cannot manage what you cannot measure&amp;rdquo;, we&amp;rsquo;d be hard-pressed to find examples of people successfully managing without measurements.&lt;/p>&lt;p>And if we could effortlessly find examples of normal people successfully managing without measurements, then it&amp;rsquo;d imply that a person not capable of managing what they cannot measure would be a muppet.&lt;/p>&lt;p>I hope it&amp;rsquo;s clear that people spend their entire life managing things they cannot measure. Some examples include:&lt;/p>&lt;ul>&lt;li>Relationships with other people, including spouses or partners, children, parents, siblings, co-workers, friends, etc.&lt;/li>&lt;li>Eating and drinking.&lt;/li>&lt;li>Hunting, fishing, or farming.&lt;/li>&lt;li>Dressing yourself according to the weather.&lt;/li>&lt;/ul>&lt;p>But I want to give a longer explanation of how nonsensical this notion that you cannot manage what you cannot measure is, to clear all doubts.&lt;/p>&lt;p>I&amp;rsquo;m originally from Italy, so it&amp;rsquo;s time to talk about pasta.&lt;/p>&lt;h3 id="the-right-amount-of-salt-in-a-pasta-dish">The &amp;ldquo;right&amp;rdquo; amount of salt in a pasta dish&lt;/h3>&lt;p>Imagine that you&amp;rsquo;re having friends for dinner, and you decide you&amp;rsquo;re going to cook some pasta, with a tomato-based fish sauce.&lt;/p>&lt;p>Pasta dishes tend to be easy, involving simple techniques, and no more than like 10 ingredients. But let&amp;rsquo;s make this even simpler, by exclusively focusing on salt. As part of the cooking, you need to add salt to your dish, with the goal of producing something you and your friends will enjoy.&lt;/p>&lt;p>If you&amp;rsquo;re from the &amp;ldquo;you cannot manage what you cannot measure&amp;rdquo; school of thought, you might believe that it&amp;rsquo;s a simple matter of measuring the right amount of salt. You scientifically determine that the right amount of salt is X grams per kilo of pasta, stick to that, and you&amp;rsquo;ll be fine.&lt;/p>&lt;p>Sounds right uh? The thing is, nobody cooks like this. Weighing the salt would be both inefficient and ineffective.&lt;/p>&lt;p>If you&amp;rsquo;re obsessed about metrics, you might claim that:&lt;/p>&lt;ol>&lt;li>Measuring is needed for mass-production, because understanding doesn&amp;rsquo;t scale.&lt;/li>&lt;li>Measuring might not be needed by the experienced cooks, but it would help the inexperienced ones.&lt;/li>&lt;/ol>&lt;p>You&amp;rsquo;d be right to say that if you want to mass-produce food, understanding doesn&amp;rsquo;t scale. You cannot mass-produce certain things, though. There&amp;rsquo;s no mass-produced pasta with tomato-based fish sauce. It&amp;rsquo;d be terrible. Software products are the same. It&amp;rsquo;s about development, not production.&lt;/p>&lt;p>And when it comes to the claim that measuring the amount of salt would help the inexperienced cook, this is totally not the case. Worse even, it&amp;rsquo;d prevent the inexperienced cook from ever learning how to cook.&lt;/p>&lt;p>First of all, no &amp;ldquo;right&amp;rdquo; amount exists, because the goal is to please people, and people have different tastes. Software product development is the same, it&amp;rsquo;s about dealing with people, needs, and preferences, not universal quantities.&lt;/p>&lt;p>Secondly, the other ingredients contribute to the saltiness of the dish, and no two ingredients are ever the same. You can never cook the same dish twice. The fish you buy, the tomatoes, they&amp;rsquo;re never the same. And the cooking process will also be different. The hob, the pots and pans, all different. So the same amount of salt will produce different results when the other ingredients differ. This is the same in software product development, and that&amp;rsquo;s why what worked elsewhere or in the past might not work in your company at present.&lt;/p>&lt;p>And, finally, salt doesn&amp;rsquo;t only contribute to the saltiness of the dish. Salt accentuates foods&amp;rsquo; own taste (its primary use). In doing so, it also makes the water contained in certain foods come out, changing the cooking process as a result. So even when you add salt is important, not just how much you add, because salting is interdependent with the cooking mechanism, so that if you want crispy onions, you shouldn&amp;rsquo;t put in any salt before they&amp;rsquo;re crispy.&lt;/p>&lt;p>Overall, even in this extremely simplified version of cooking a simple dish, there are complexities and nonlinear effects that make a metrics-based approach unfeasible. And yet most people can consistently manage to cook a good dish, without measuring the amount of salt (or anything else).&lt;/p>&lt;p>And this means that not only you can manage what you cannot measure, but also that if you cannot manage what you cannot measure, then you wouldn&amp;rsquo;t be able to manage it even if you had those measurements.&lt;/p>&lt;h2 id="its-about-understanding-not-metrics">It&amp;rsquo;s about understanding, not metrics&lt;/h2>&lt;p>The reason why people can manage without relying on metrics is because the key to managing well is understanding, not data.&lt;/p>&lt;p>Understanding produces principles about why things work the way they work, enabling us to orient ourselves when dealing with uncertainty.&lt;/p>&lt;h2 id="back-to-software-product-development">Back to software product development&lt;/h2>&lt;p>Software product development is incredibly more complex than cooking some pasta.&lt;/p>&lt;p>There are entire dimensions of added complexity, including:&lt;/p>&lt;ol>&lt;li>It&amp;rsquo;s done in teams, working together.&lt;ul>&lt;li>This is not the case, even in professional kitchens, as people split the work between them, rather than collaborating on the same dish.&lt;/li>&lt;/ul>&lt;/li>&lt;li>The subject is far removed from the realm of common sense and experiences.&lt;ul>&lt;li>The untrained person knows next to nothing about it.&lt;/li>&lt;li>Good and bad products and software are not immediately recognizable, unlike food.&lt;/li>&lt;/ul>&lt;/li>&lt;li>It&amp;rsquo;s much longer term.&lt;ul>&lt;li>You don&amp;rsquo;t often get to start from scratch&lt;/li>&lt;li>Software products and systems get continuously evolved and adapted to new requirements.&lt;/li>&lt;li>What you did in the past influences and limits what you can do in the future.&lt;/li>&lt;/ul>&lt;/li>&lt;li>The landscape changes continuously.&lt;ul>&lt;li>After tomatoes made their way to Europe, the recipe for pasta with fish and tomato sauce stayed virtually unchanged. Even the tools are roughly the same.&lt;/li>&lt;li>In software, things change continuously. Entire paradigms appear and force everybody to shift. Cloud providers, containers, orchestrators, infrastructure as code, etc.&lt;/li>&lt;/ul>&lt;/li>&lt;li>The scale of the system is much larger.&lt;ul>&lt;li>The recipe for pasta with fish and tomato sauce might involve 10 ingredients or so.&lt;/li>&lt;li>A software product might involve 10s or 1000s of systems, features, and moving parts.&lt;/li>&lt;/ul>&lt;/li>&lt;/ol>&lt;h2 id="final-words">Final words&lt;/h2>&lt;p>It turns out that you can indeed manage what you cannot measure, and do it all the time.&lt;/p>&lt;p>So shall we stop misusing this quote to push this craziness for metrics further, uh?&lt;/p></description></item><item><title>Make the change easy, then make the easy change</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-30-make-the-easy-change/</link><pubDate>Sat, 30 Sep 2023 05:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-30-make-the-easy-change/</guid><description>&lt;p>&amp;ldquo;Make the change easy, then make the easy change&amp;rdquo;. Didn&amp;rsquo;t Kent Beck say that? He did, didn&amp;rsquo;t he? So why are we talking about this (again)? I just want to clarify a few things about this very good piece of advice.&lt;/p>&lt;h2 id="the-example">The example&lt;/h2>&lt;p>To show what I mean, I&amp;rsquo;m going to use a fictional example.&lt;/p>&lt;p>Imagine you&amp;rsquo;re a software developer working for an upcoming startup that wants to dominate the online clothes business.&lt;/p>&lt;p>You&amp;rsquo;re working on the MVP, and somebody decided that there&amp;rsquo;s only going to be 1 T-shirt available, in 1 size and 1 colour, and the price is going to be $7. More precisely, you&amp;rsquo;ll only be able to order a pair: 2 t-shirts for $14.&lt;/p>&lt;h2 id="when-yagni-goes-wrong">When YAGNI goes wrong&lt;/h2>&lt;p>You&amp;rsquo;re a good developer, so you start thinking, but then you remember you read something about YAGNI. Why don&amp;rsquo;t you just build what&amp;rsquo;s needed for now?&lt;/p>&lt;p>After all why bother doing more than what&amp;rsquo;s strictly necessary?&lt;/p>&lt;p>You create an HTTP endpoint that accepts a &lt;code>POST&lt;/code> on &lt;code>/orders&lt;/code>, without any further order-related information, and you process each of them by charging $14 and by initiating the delivery of 2 t-shirts of the 1 size and colour you support. You add a test (before or after), and you&amp;rsquo;re done.&lt;/p>&lt;p>You haven&amp;rsquo;t wasted any time making choices explicit, modelling your domain, writing serialization logic, making things modular and decoupled, reasoning about where the information comes from, or making bad decisions about using doubles for money amounts.&lt;/p>&lt;p>It feels pretty good.&lt;/p>&lt;p>You delivered on the requirements, made the customers happy, and saved time you could have spent gold-plating the implementation. Aced it, OKRs met, mission accomplished. You got closer to that promotion for sure.&lt;/p>&lt;h2 id="the-problem">The problem&lt;/h2>&lt;p>By not identifying circumstantial conditions and temporary constraints you created a liability for your organisation.&lt;/p>&lt;p>But could we not evolve the code later, when needed? Isn&amp;rsquo;t Agile about delaying decisions after all?&lt;/p>&lt;p>Agile is about delaying decisions until the latest responsible moment. Here we&amp;rsquo;re way past that point.&lt;/p>&lt;p>The problem is that the cost of doing things changes during the lifecycle of a project. And your employers aren&amp;rsquo;t aware that you traded off a few hours of additional work for a much higher risk and effort down the line.&lt;/p>&lt;p>With any luck, the requirements will evolve quickly enough that you&amp;rsquo;ll feel naive, you&amp;rsquo;ll undo the damage, but nothing too bad will happen.&lt;/p>&lt;p>In a much less forgiving scenario, your company will thrive for quite a while before requiring changes that your implementation doesn&amp;rsquo;t support. Imagine going on for a year or so under the same requirements. You build UIs, logging, monitoring, metrics, alerts, an inventory system, and more. All while keeping everything implicit as above.&lt;/p>&lt;p>When your requirements will finally expand, the effort, risk, and cost will be obscene. If your company will need those changes quickly, it might be a big problem. The company might not even survive.&lt;/p>&lt;h2 id="yagni-done-right">YAGNI done right&lt;/h2>&lt;p>What&amp;rsquo;s the alternative then? It&amp;rsquo;s YAGNI, but done right.&lt;/p>&lt;p>You model enough, without imagining future requirements that you ain&amp;rsquo;t going to need.&lt;/p>&lt;p>But isn&amp;rsquo;t this what we did above?&lt;/p>&lt;p>Nope. We need a distinction. YAGNI is about behaviour, not quality.&lt;/p>&lt;p>The problem with the approach above is that, by modelling exactly accordingly to the current requirements, you leave information out of your model. When your requirements evolve, you&amp;rsquo;ll not only need to fix the logic, but you&amp;rsquo;ll be forced to change everything to get the information you need for the decision making.&lt;/p>&lt;p>What do you do for the scenario above instead then?&lt;/p>&lt;ol>&lt;li>You create a product ID for the 1 t-shirt you &amp;ldquo;currently&amp;rdquo; only sell.&lt;/li>&lt;li>You create the concept of an order, supporting multiple items and quantities.&lt;/li>&lt;li>You introduce the concept of product options, to allow sizes and colours.&lt;/li>&lt;li>You adopt a money amount type to model prices and costs.&lt;/li>&lt;li>You abstract pricing an order behind an interface.&lt;/li>&lt;li>You abstract checking whether an order is valid behind an interface.&lt;/li>&lt;li>You create an endpoint to accept orders.&lt;/li>&lt;li>You take care of serialising and deserialising your models.&lt;/li>&lt;li>You create modules and layers, and test these in isolation.&lt;/li>&lt;/ol>&lt;p>And then, for the time being:&lt;/p>&lt;ol>&lt;li>You implement the order validator in a way that endures the only product available is your t-shirt, in that 1 size and colour, and that the quantity ordered is exactly 2.&lt;/li>&lt;li>You return $14 as a money amount type from the order pricing implementation.&lt;/li>&lt;li>For now all the logic might even stay in the code, as opposed to living in a database (price, existing products, existing product options, order constraints) (debatable, as changing these will require a code change an a release, but could work for a while).&lt;/li>&lt;/ol>&lt;p>When your company&amp;rsquo;s requirements will change, you&amp;rsquo;ll evolve your pricing or validation implementation, without having to change the whole codebase to propagate the information you didn&amp;rsquo;t use to need but you now need.&lt;/p>&lt;p>This is typically easy and painless, because you made the change easy, and then you made the easy change.&lt;/p>&lt;h2 id="final-words">Final words&lt;/h2>&lt;p>It&amp;rsquo;s important to separate speculating about future behaviour from modelling your domain without removing information you might not currently need, but that is part of the decision in the general case.&lt;/p>&lt;p>So if the cost of an order depend on the order, the time, and the shopper, the abstraction that models the pricing should receive this information, even if at the moment it doesn&amp;rsquo;t use it.&lt;/p>&lt;p>This applies equally well to domain events. You include all the information you have, not what you need for the next processing step.&lt;/p>&lt;p>And, with that said, I&amp;rsquo;m done. I hope this will make you think, even if you end up disagreeing with me.&lt;/p></description></item><item><title>It's time to put REST to rest</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-22-put-rest-to-rest/</link><pubDate>Fri, 22 Sep 2023 07:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-22-put-rest-to-rest/</guid><description>&lt;p>REST is still going strong these days, with many companies structuring their HTTP API based on its principles. And yet, I think that REST is a fundamental mistake, that it caused a lot of damage and pain, and that people should just stop creating RESTful HTTP APIs.&lt;/p>&lt;p>A bold statement, isn&amp;rsquo;t it? Let&amp;rsquo;s dig a little bit deeper to understand the problem, and an alternative.&lt;/p>&lt;h1 id="what-is-rest-a-recap">What is REST: a recap&lt;/h1>&lt;p>I won&amp;rsquo;t explain REST in details, but a brief and incomplete summary is that, in a RESTful HTTP API:&lt;/p>&lt;ol>&lt;li>Users manipulate resources.&lt;/li>&lt;li>Resources have a representation in various data formats e.g. JSON, but even XML or others.&lt;/li>&lt;li>A resource is associated to a specific request path.&lt;/li>&lt;li>Resources can be nested to model relationships.&lt;/li>&lt;li>Some HTTP methods assume a specific meaning, representing particular actions a user can do on a resource:&lt;ul>&lt;li>&lt;code>GET&lt;/code> retrieves a resource e.g. &lt;code>GET /schools/{school-ID}/students/{student-ID}&lt;/code> would retrieve a student with given ID of the school with given ID. The response body might look like &lt;code>{ &amp;quot;id&amp;quot;: &amp;quot;123&amp;quot;, &amp;quot;firstname&amp;quot;: &amp;quot;Bruce&amp;quot;, &amp;quot;lastname&amp;quot;: &amp;quot;Wayne&amp;quot;}&lt;/code>&lt;/li>&lt;li>&lt;code>POST&lt;/code> creates a new resource e.g. &lt;code>POST /schools/{school-ID}/students { &amp;quot;firstname&amp;quot;: &amp;quot;Peter&amp;quot;, &amp;quot;lastname&amp;quot;: &amp;quot;Parker&amp;quot; }&lt;/code>&lt;/li>&lt;li>&lt;code>DELETE&lt;/code> removes a resource e.g. &lt;code>DELETE /schools/{school-ID}/students/123&lt;/code>&lt;/li>&lt;li>&lt;code>PUT&lt;/code> replaces a resource in its entirety e.g. &lt;code>PUT /schools/{school-ID}/students/234 { &amp;quot;firstname&amp;quot;: &amp;quot;Clark&amp;quot;, &amp;quot;lastname&amp;quot;: &amp;quot;Kent&amp;quot; }&lt;/code>&lt;/li>&lt;li>&lt;code>PATCH&lt;/code> replaces a field in a resource with a new value &lt;code>PATCH /schools/{school-ID}/students/234 { &amp;quot;lastname&amp;quot;: &amp;quot;Luthor&amp;quot; }&lt;/code>&lt;/li>&lt;/ul>&lt;/li>&lt;/ol>&lt;h2 id="the-benefits-rest-is-supposed-to-introduce">The &amp;ldquo;benefits&amp;rdquo; REST is supposed to introduce&lt;/h2>&lt;p>Some folks claim that REST introduces the following benefits:&lt;/p>&lt;ol>&lt;li>It&amp;rsquo;s simple, and easy to understand. You can perform the same actions on all resources, so you don&amp;rsquo;t need to learn the details of a specific API to use it.&lt;/li>&lt;li>The API is usable via a browser.&lt;/li>&lt;/ol>&lt;h2 id="i-dont-lack-experience-with-it">I don&amp;rsquo;t lack experience with it&lt;/h2>&lt;p>First of all, a brief recap of my experience here. I started my back-end development career in a company that was using SOAP as its API style. I quickly graduated to REST, and then went on to use it for 12 years. My job has primarily been developing and evolving software products that use HTTP APIs extensively.&lt;/p>&lt;p>I was a big REST enthusiast, and I even used HATEOAS principles when building HTTP APIs. I came to question the whole REST idea only recently.&lt;/p>&lt;p>This doesn&amp;rsquo;t mean you need to take my word for it, I&amp;rsquo;m just saying that this is not about me being unfamiliar with REST.&lt;/p>&lt;h1 id="the-problem-with-rest">The problem with REST?&lt;/h1>&lt;p>So what&amp;rsquo;s the problem with REST anyway?&lt;/p>&lt;h2 id="its-benefits-are-bollocks">Its benefits are bollocks&lt;/h2>&lt;p>First of all, I want to spend a few words about the benefits that REST is supposed to introduce.&lt;/p>&lt;ol>&lt;li>About being able to use any API without understanding it first: you cannot perform the same actions on all resources, because those actions have business meaning. This is the central argument against REST, so more on this below, but you cannot use any RESTful API immediately just because it&amp;rsquo;s RESTful.&lt;/li>&lt;li>About being able to use a RESTful API directly from a browser: browsers only use the &lt;code>GET&lt;/code> and &lt;code>POST&lt;/code> HTTP methods, among the ones REST associates meaning to. You cannot use a RESTful API from the browser, because the &lt;code>PUT&lt;/code>, &lt;code>PATCH&lt;/code>, and &lt;code>DELETE&lt;/code> methods are not used by the browser, and even the &lt;code>POST&lt;/code> method cannot be used without a UI. Also, even if you could, APIs are for machines, not humans. The user experience would be dreadful if you had to use a RESTful API directly from a browser, without a specific UI for it.&lt;/li>&lt;/ol>&lt;h2 id="the-fundamental-issue">The fundamental issue&lt;/h2>&lt;p>The fundamental issue with REST is that it defines operations that manipulate data structures. It&amp;rsquo;s about structure and structural changes, not behaviour. It&amp;rsquo;s imperative, rather than declarative.&lt;/p>&lt;p>Sounds like a pile of philosophical nonsense to you? Well, it&amp;rsquo;s about to get a lot more concrete.&lt;/p>&lt;p>The problem with this approach is that software products aren&amp;rsquo;t about arbitrary JSON (or XML or whatever) data structures being manipulated. Unless your product stores arbitrary digital records, that is.&lt;/p>&lt;p>So if your business deals with schools and students, it&amp;rsquo;s going to have business-specific operations that have their own meaning.&lt;/p>&lt;p>You don&amp;rsquo;t create a new student under a school, you enlist them. And if a student decides to leave your school, they don&amp;rsquo;t get deleted, a record remains. Replacing a student with another makes no sense whatsoever.&lt;/p>&lt;p>See what&amp;rsquo;s happening here? &lt;strong>REST replaces business-specific operations with generic changes to the structure of some resources&lt;/strong>.&lt;/p>&lt;h2 id="the-implications">The implications&lt;/h2>&lt;p>Is it really that bad though? I mean, sure, it&amp;rsquo;s a bit clunky, but does it really matter? As a matter of fact, it does.&lt;/p>&lt;p>By stripping away the meaning of the business operation, REST guarantees that a RESTful HTTP API will be hard to understand, to use, and to maintain.&lt;/p>&lt;p>If you understand trading, you might look for a way to place a limit order on an asset, or a market order. The action that resonates with you is &amp;ldquo;place&amp;rdquo;, not &amp;ldquo;post&amp;rdquo;.&lt;/p>&lt;p>There might be multiple operations that result in the creation or deletion of the same data, and REST doesn&amp;rsquo;t support this very well (or at all, really).&lt;/p>&lt;p>By expressing business operations as data manipulations, REST makes validation a pain. When a user clicks a button, there&amp;rsquo;s a clear business operation that they&amp;rsquo;re performing. Then the client strips away its meaning, by converting it to a RESTful API call. On the receiving side, the server needs to reconstruct the meaning of the original business operation, to process it correctly. Because let&amp;rsquo;s face it, no company (I hope!) blindly applies data manipulations to a record in a database.&lt;/p>&lt;p>So when the server receives a &lt;code>PATCH /schools/{school-ID}/students/234 { &amp;quot;lastname&amp;quot;: &amp;quot;Luthor&amp;quot; }&lt;/code> request, it needs to understand that a student has requested for their lastname to be changed, and it then needs to decide what to do about it. What if some fields cannot be changed? What if the value of some fields is constrained by the value of some other fields? By allowing an arbitrary &lt;code>PATCH&lt;/code> operation, these concepts are hard to model and validate requests against.&lt;/p>&lt;p>No company will allow to replace a resource with another, because it makes little sense from a business perspective.&lt;/p>&lt;p>Also, a RESTful API is forced to represent the same business concepts with the same structure across all its operations. This can be quite inconvenient.&lt;/p>&lt;p>And if you rename your business concepts, you need to version your API.&lt;/p>&lt;p>What about the way REST advocates to use HTTP status codes? So when a student with given ID doesn&amp;rsquo;t exist, a RESTful API returns the &lt;code>404&lt;/code> HTTP status code.&lt;/p>&lt;p>This also sucks, because HTTP status codes are about the HTTP protocol, not your business semantics. The &lt;code>404 Not Found&lt;/code> status code indicates that a path doesn&amp;rsquo;t exist. If you use it to say that a student doesn&amp;rsquo;t exist, these two get mixed up. If a client invokes the wrong path by mistake, there&amp;rsquo;s no telling whether it&amp;rsquo;s due to a protocol error or to a nonexistent student.&lt;/p>&lt;h1 id="the-alternative">The alternative&lt;/h1>&lt;p>So what can we do instead? Should we go back to SOAP, with its verbosity? Well, not really. And I&amp;rsquo;m not talking about &lt;code>gRPC&lt;/code> either.&lt;/p>&lt;p>The idea is to model business-specific operations, and to allow users to invoke these, instead of generic structure-related data manipulations.&lt;/p>&lt;p>The resulting HTTP API tends to always look the same:&lt;/p>&lt;p>&lt;code>POST&lt;/code> on &lt;code>/commands/{command-type}/version/{command-version}&lt;/code> and, if you prefer to keep them semantically separated, &lt;code>POST&lt;/code> on &lt;code>/queries/{query-type}/version/{query-version}&lt;/code>.&lt;/p>&lt;p>The request body defines the data, using a structure that&amp;rsquo;s specific to the command/query type and version. Basically each command/query is its own API, meaning that if you rename your business concepts, all you need to do is introduce additional commands/queries and deprecate the old ones, without having to version your whole API e.g. the request paths.&lt;/p>&lt;p>Going back to our school example above, you might end up with the following operations:&lt;/p>&lt;ol>&lt;li>&lt;code>POST&lt;/code> on &lt;code>/commands/enlist-student/version/1 { &amp;quot;firstname&amp;quot;: &amp;quot;Bruce&amp;quot;, &amp;quot;lastname&amp;quot;: &amp;quot;Wayne&amp;quot;, &amp;quot;joining-date&amp;quot;: &amp;quot;2023-09-22&amp;quot; }&lt;/code> to enlist a student. Want to operate across multiple students or schools? Just accept multiple fields as part of a dedicated command.&lt;/li>&lt;li>&lt;code>POST&lt;/code> on &lt;code>/queries/enlisted-students-on-joining-date/version/1 { &amp;quot;date&amp;quot;: &amp;quot;2023-09-22&amp;quot; }&lt;/code> to retrieve all students that joined on a given date.&lt;/li>&lt;li>&lt;code>POST&lt;/code> on &lt;code>/commands/report-student-lastname-change/version/1 { &amp;quot;student-id&amp;quot;: &amp;quot;123&amp;quot;, &amp;quot;new-lastname&amp;quot;: &amp;quot;Kent&amp;quot; }&lt;/code> to report a lastname change for a student.&lt;/li>&lt;/ol>&lt;p>See what&amp;rsquo;s happening here? The API is way easier to understand, document, use, and maintain. The server can route and process based on command/query type and version. You can bind a response description for a path in your OpenAPI specification easily. OpenAPI doesn&amp;rsquo;t allow if statements, so an arbitrary REST-style &lt;code>PATCH&lt;/code> is hard to document.&lt;/p>&lt;p>And you get batching capabilities for free:&lt;/p>&lt;p>&lt;code>POST&lt;/code> on &lt;code>/commands { &amp;quot;commands&amp;quot;: [ { &amp;quot;type&amp;quot;: &amp;quot;enlist-student&amp;quot;, &amp;quot;version&amp;quot;: 1, &amp;quot;data&amp;quot;: { &amp;quot;firstname&amp;quot;: &amp;quot;Bruce&amp;quot;, &amp;quot;lastname&amp;quot;: &amp;quot;Wayne&amp;quot;, &amp;quot;joining-date&amp;quot;: &amp;quot;2023-09-22&amp;quot; } }, { &amp;quot;type&amp;quot;: &amp;quot;report-student-lastname-change&amp;quot;, &amp;quot;version&amp;quot;: 1, &amp;quot;data&amp;quot;: { &amp;quot;student-id&amp;quot;: &amp;quot;123&amp;quot;, &amp;quot;new-lastname&amp;quot;: &amp;quot;Kent&amp;quot; } } ] }&lt;/code>.&lt;/p>&lt;p>Batching commands and/or queries can have tremendous throughput implications, when applicable.&lt;/p>&lt;p>Last, your semantics can better differentiate between HTTP protocol errors and application level results.&lt;/p>&lt;p>When a requested student is nonexistent, your API can return &lt;code>200 OK&lt;/code> HTTP status code with a &lt;code>{ &amp;quot;user&amp;quot;: null, &amp;quot;message&amp;quot;: &amp;quot;No user exists with the specified ID&amp;quot; }&lt;/code> response body.&lt;/p>&lt;h1 id="parting-words">Parting words&lt;/h1>&lt;p>That&amp;rsquo;s it really: &lt;strong>REST replaces business-specific operations with generic changes to the structure of some resources, and that is evil&lt;/strong>.&lt;/p>&lt;p>I hope I was able to clearly explain why this is a terrible idea, and what you could and should do instead. I don&amp;rsquo;t hate &lt;code>gRPC&lt;/code> by the way, but I don&amp;rsquo;t think it&amp;rsquo;s a good approach for client to server communications.&lt;/p>&lt;p>So next time you need to design an HTTP API, why don&amp;rsquo;t you stick to business-specific operations and make your users&amp;rsquo; life and your own life simpler?&lt;/p></description></item><item><title>Why software development is hard</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-12-why-software-development-is-hard/</link><pubDate>Tue, 12 Sep 2023 11:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-12-why-software-development-is-hard/</guid><description>&lt;p>We always hear that software development is hard, and that a very high percentage of all software projects fail. But have you ever wondered why software development is so hard?&lt;/p>&lt;p>I&amp;rsquo;m not considering software product development as a whole here, merely its software development aspects. Product development adds many other dimensions of complexity to software development, and I don&amp;rsquo;t want to mix the two.&lt;/p>&lt;h1 id="causes-of-complexity">Causes of complexity&lt;/h1>&lt;p>Let&amp;rsquo;s look at some causes for this complexity.&lt;/p>&lt;h2 id="misinformation-abundance">(Mis)information abundance&lt;/h2>&lt;p>Software development should be well understood by now. After all, we&amp;rsquo;ve been struggling with it for 50+ years at least, and there&amp;rsquo;s a wealth of information about it.&lt;/p>&lt;p>You can read books, listen to podcasts, or watch videos, and there are classes and courses. You can often access all of this for free, or without breaking the bank in the worst case.&lt;/p>&lt;p>And if it&amp;rsquo;s hired help you&amp;rsquo;re looking for, you&amp;rsquo;d hardly struggle to find it: consultancies, contractors, advisors, coaches, and experts are at easy reach for the executive in need.&lt;/p>&lt;p>And yet, both doing the right thing and doing the thing right remain elusive. Despite all this information available, it&amp;rsquo;s hard to know what to do, what not to do, and how to do things.&lt;/p>&lt;p>Books, courses, and other mechanisms to access information provide an analysis of the subject they concern themselves with. They break down the subject into its parts, and then analyze the parts, yielding knowledge of how the subject they treat works. Analysis never yields understanding (&amp;ldquo;why something works the way it works&amp;rdquo;), or wisdom (&amp;ldquo;what to do or not to do about something&amp;rdquo;).&lt;/p>&lt;h2 id="a-fast-moving-industry">A fast moving industry&lt;/h2>&lt;p>The software development industry moves fast. Trends, frameworks, languages, open-source projects, security vulnerabilities, databases, build tools, packaging, distribution, Cloud services: they all come, change, and go fast.&lt;/p>&lt;p>It&amp;rsquo;s not about keeping up with the latest fad. Technologies change all the time, but principles tend to remain the same. Sometimes, though, new technologies dramatically simplify or enhance something, leading to a paradigm shift.&lt;/p>&lt;p>When this happens, companies that don&amp;rsquo;t adapt quickly find themselves at serious disadvantage compared to those that do. And just keeping track of what&amp;rsquo;s new and what changed is a huge effort.&lt;/p>&lt;h2 id="increasing-expectations">Increasing expectations&lt;/h2>&lt;p>Open-source technologies and Cloud providers have enabled software developers to achieve things that 30 years ago were impossible. And yet, this has raised the bar in terms of expectations on the software itself.&lt;/p>&lt;p>Distributed systems, virtualization, containers, orchestration, zero-downtime deployments, security vulnerability scanning, dependency management, runtime agents, upgrades, auditability, service mesh, responsive and accessible UIs, push notifications, horizontal and vertical scaling, idempotency, data residency regulations, multi-region availability, vendor-lock in, multi-Cloud, eventual consistency, encryption, GDPR, etc.&lt;/p>&lt;p>Layers over layers of requirements, expectations, integrations, and accidental complexity. This complexity, while accidental because not intrinsic to the business operations implemented by the software, is often unavoidable, because customer expectations and regulations are requirements the software must meet.&lt;/p>&lt;h2 id="invisible-quality-and-demand">Invisible quality and demand&lt;/h2>&lt;p>The whole manufacturing industry is about demand and quality. You predict the variety of demand by looking at historical data, and you gear yourself up to meet this demand. Quality can be defined by how little you deviate from expectations. So if something is supposed to weight 50g, 49g and 51g are equally bad.&lt;/p>&lt;p>Service organizations can still look at historical data to predict demand, but they struggle with defining quality. A barber could record the frequency of hair types, requested cuts, and so on, but they&amp;rsquo;d struggle to measure the quality of their work. This is because, for service providers, quality is in the eye of the beholder. You cannot measure the quality of a barber, of a lawyer, or of a city planner.&lt;/p>&lt;p>For software development companies, unfortunately, even demand is an elusive concept. It might work for software development agencies that work on short contracts. These behave in a pretty similar way to a barber. But most companies aren&amp;rsquo;t about short engagements. For most companies, at any given time, it&amp;rsquo;s impossible to know what demand they&amp;rsquo;ll experience, what quality they provide to their customers, or even what quality they have internally.&lt;/p>&lt;h2 id="collaborative-knowledge-work">Collaborative knowledge work&lt;/h2>&lt;p>Software development is collaborative knowledge work. You have people with different strengths, experience, knowledge, roles, desires, agendas, and personalities. And they need to work together in a team, while often competing with each other over limited resources e.g. money, status, promotions, or attention.&lt;/p>&lt;p>People are systems, and their performance is hugely influenced by their state of mind, including focus, mood, psychological safety, and by how their needs are met by their job, team, company, industry, city, and personal life.&lt;/p>&lt;p>Teams are also systems, meaning that their performance is never the sum of the parts, but the product of the interactions between the team members, which are never interchangeable.&lt;/p>&lt;p>Companies are systems as well, and evolve continuously in terms of their employees, their position in the market, the level of resources they command, and their underlying view of the world, culture, and operating principles.&lt;/p>&lt;h2 id="continuously-evolving-the-same-piece">Continuously evolving the same piece&lt;/h2>&lt;p>Software development is not manufacturing. The teams don&amp;rsquo;t build or produce anything. That&amp;rsquo;s what the build pipelines are for. Teams develop software. Like developing a vaccine, a recipe, or an invention.&lt;/p>&lt;p>You continuously work on your software system, evolving it to meet changing and expanding requirements. And it&amp;rsquo;s not a linear process either:&lt;/p>&lt;ul>&lt;li>Inputs come as requirements.&lt;/li>&lt;li>The system evolves to meet these additional requirements, while also meeting all the previous ones.&lt;/li>&lt;li>The new capabilities developed inform the future inputs, with expanding scope or rework.&lt;/li>&lt;/ul>&lt;h1 id="to-sum-it-up">To sum it up&lt;/h1>&lt;p>Overall, I believe the reason why software development is so hard is that it&amp;rsquo;s about teams of people developing and continuously evolving software systems, to provide a service to their organization.&lt;/p>&lt;p>From this you get:&lt;/p>&lt;ul>&lt;li>Inability to quantify most things.&lt;/li>&lt;li>All the complexities of systems: balancing and reinforcing loops, power laws, potential instability, higher order causes, change-resistant behavior.&lt;/li>&lt;li>The complexities of social systems: the differences between people, the need to cooperate, the effect of incentives, the internal competition, the changes in management, hierarchy, market position, employees, worldview, culture, etc.&lt;/li>&lt;li>The complexities of distributed software systems: technologies, requirements, integrations, vendors, regulations, trends, and threats: all ever-changing and moving fast.&lt;/li>&lt;li>The interactions between social systems and software systems: Conway&amp;rsquo;s Law, the effects that the quality of the codebase has on the people working on it, etc.&lt;/li>&lt;li>The complexities of long-term horizons, and having to deal with the consequences of your decisions for a long time, with compounding effects.&lt;/li>&lt;/ul>&lt;h1 id="parting-words">Parting words&lt;/h1>&lt;p>Before ending this post, a few final considerations:&lt;/p>&lt;ul>&lt;li>In a complex scenario, best practices don&amp;rsquo;t make sense. You never face the same situation, across either projects or companies. You probe, sense, and act. That&amp;rsquo;s why there&amp;rsquo;s so much information out there, and why it&amp;rsquo;s mostly so useless.&lt;/li>&lt;li>The internal quality of a software system determines the complexity and the effort involved in changing the system itself. The complexity and effort involved in an activity, in turn, determine the likelihood that a team might perform this activity. So bad systems are hard to change, require more work to be operated, and tend to get worse over time, as people pile up shortcuts over shortcuts.&lt;/li>&lt;li>Software development is a continuous fight against entropy. Quality is the only way to achieve speed. Anything that destroys the pride a craftsman takes in their work is a major danger to any software project.&lt;/li>&lt;li>The wrong incentives will favor internal competition and defensiveness, while destroying cooperation. This will also manifest itself in the processes and in the software systems of a company.&lt;/li>&lt;/ul></description></item><item><title>For crying in the beer, stop writing unit tests!</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-07-stop-writing-unit-tests/</link><pubDate>Thu, 07 Sep 2023 11:30:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-09-07-stop-writing-unit-tests/</guid><description>&lt;p>Stop writing unit tests! It&amp;rsquo;s wrong on many levels, so don&amp;rsquo;t do it. Actually remove the ones you have.&lt;/p>&lt;p>I&amp;rsquo;m a huge fan of automated tests, and I use Test-Driven Development a lot, so no, I&amp;rsquo;m not advocating against testing in general. And don&amp;rsquo;t think the solution is to write a lot of end-to-end tests either, as that&amp;rsquo;s even worse.&lt;/p>&lt;p>So what&amp;rsquo;s wrong with unit tests? And what should we do instead? Let&amp;rsquo;s have a look&amp;hellip;&lt;/p>&lt;h2 id="how-most-developers-approach-testing">How most developers approach testing&lt;/h2>&lt;p>Assuming they write tests at all, people tend to write unit, integration, and end-to-end tests.&lt;/p>&lt;p>Leaving aside the names for a second, as there&amp;rsquo;s no consensus on these, the idea is that there are units, like lego blocks, and these units interact to deliver functionality as part of a service. Services, in turn, interact as part of a system.&lt;/p>&lt;p>With this in mind, the usual test strategy involves the following test types.&lt;/p>&lt;h3 id="unit-tests">Unit tests&lt;/h3>&lt;p>Unit tests focus on your units. In particular, a unit test covers a unit in isolation from other units. What&amp;rsquo;s a unit? Well, a unit is a function or type e.g. a class.&lt;/p>&lt;p>You construct the unit, isolate your test by replacing its dependencies with mocks or stubs you can control, then excite the unit, and check the outputs and the internal state.&lt;/p>&lt;p>Examples:&lt;/p>&lt;ul>&lt;li>You have a car with 4 wheels, you remove 1 wheel, and you check that the car now has 3 wheels.&lt;/li>&lt;li>You create a car of a make and colour, and you check that its colour and make match the ones you passed as inputs.&lt;/li>&lt;li>You have a 1m-high table, and you check its height property is equal to 1m.&lt;/li>&lt;/ul>&lt;p>I can see some of you screaming, but this is a popular approach. I&amp;rsquo;m aware that this is not what Kent Beck meant, more on this towards the end of the post.&lt;/p>&lt;h3 id="integration-tests">Integration tests&lt;/h3>&lt;p>Integration tests, as the name suggests, focus on testing the integration logic. So what are we integrating and testing? Well, we&amp;rsquo;re integrating various units together, and we&amp;rsquo;re also integrating with databases, etc.&lt;/p>&lt;p>Now that we have tested the lego blocks, we cover their integration with other blocks or downstream dependencies.&lt;/p>&lt;p>How does this work? You spin up your service, create some preconditions inside your downstream dependencies e.g. databases or queues, you excite your service using its API e.g. through an HTTP client, and check the outputs and the internal state in the DB or queue after the changes. If something is misconfigured at service level, integration tests have good chances of catching it.&lt;/p>&lt;p>These tests catch more issues than unit tests, but are slower. Also, expectations are harder to set up at database or queue level, so many conditions are harder to simulate at this level (e.g. delays, dependencies not being available or misbehaving). So we don&amp;rsquo;t just rely on these, but also on unit tests.&lt;/p>&lt;h3 id="end-to-end-tests">End-to-end tests&lt;/h3>&lt;p>End-to-end tests cover the interaction between different services and systems. They go from one end to the opposite end, are driven by the input closest to the user e.g. a UI or a top-level API, and they drive all sorts of upstream and downstream systems, including load balancers, databases, Cloud providers, and even SaaS vendors.&lt;/p>&lt;p>These are slow, cumbersome, often unstable, poorly isolated, and require a full environment to run, but hopefully we don&amp;rsquo;t need as many, once we have unit and integration tests in place.&lt;/p>&lt;h2 id="doesnt-this-work">Doesn&amp;rsquo;t this work?&lt;/h2>&lt;p>So what&amp;rsquo;s the problem with this approach? It sounds sensible after all. In order to understand the problem, we need to think about what great looks like.&lt;/p>&lt;h3 id="what-great-looks-like">What great looks like&lt;/h3>&lt;p>A great automated test suite is surprisingly intuitive: it provides you with quick and accurate confidence that your changes are good to go.&lt;/p>&lt;p>The feedback must be quick. Speed really matters here. Delay the feedback, and you get all sorts of problems, including people switching to other tasks while waiting, increasing work in progress, rework, etc.&lt;/p>&lt;p>The feedback must also be accurate. This means you should be able to pinpoint the problem whenever there are test failures.&lt;/p>&lt;p>What does confidence mean here though? Everybody immediately recognises that confidence means that when things are wrong, your tests should indeed tell you that things are wrong. So false positives are bad.&lt;/p>&lt;p>Is this enough though? Well, no, it isn&amp;rsquo;t. You also need confidence that when things are working, your test suite should tell you that things are okay. False negatives are a real bummer as well.&lt;/p>&lt;h3 id="the-problem">The problem&lt;/h3>&lt;p>There are so many problems with the unit, integration, and end-to-end test strategy, and they all fundamentally stem from a misconception of how systems work.&lt;/p>&lt;p>Like Ackoff used to say, &amp;ldquo;A system is never the sum of its parts; it’s the product of their interaction&amp;rdquo; and &amp;ldquo;The performance of a system doesn&amp;rsquo;t depend on how the parts perform taken separately, it depends on how they perform together – how they interact, not on how they act, taken separately&amp;rdquo;.&lt;/p>&lt;p>The thinking behind this popular test strategy is that your software system is the sum of its parts, or units. So if the units are working correctly, then the system is also going to work well.&lt;/p>&lt;p>I can hear you thinking &amp;ldquo;uh, not so, because we have integration and end-to-end tests, so we care about the interactions as well&amp;rdquo;. True, but there&amp;rsquo;s nuance to this.&lt;/p>&lt;p>By thinking of a system as the sum of its parts, we can&amp;rsquo;t let go of the need for tests that focus on the parts in isolation: our unit tests. After all, if a part is not working well, how could its containing system ever work well, right? And what&amp;rsquo;s the issue with unit tests anyway?&lt;/p>&lt;p>When unit tests are defined as above, focusing on classes or functions in isolation, they cannot test behaviour, because behaviour is a property of a system, never of its parts. A class doesn&amp;rsquo;t do anything in isolation; behaviour comes from the interactions.&lt;/p>&lt;p>So if they cannot test behaviour, what do unit tests test? Structure. Let&amp;rsquo;s revisit the examples above:&lt;/p>&lt;ul>&lt;li>You have a car with 4 wheels, you remove 1 wheel, and you check the car now has 3 wheels.&lt;/li>&lt;li>You create a car of a make and colour, and you check that its colour and make match the ones you passed as inputs.&lt;/li>&lt;li>You have a 1m-high table, and you check its height property is 1m.&lt;/li>&lt;/ul>&lt;p>None of these tests is about behaviour. They test structure, e.g. state or properties. That there&amp;rsquo;s a table, and that its height is 1m. That a car has 4 wheels.&lt;/p>&lt;p>The problem here is that you care about the correct behaviour of your system, not about its structure.&lt;/p>&lt;p>You might counter with the argument that, at worst, unit tests are not that useful, but I believe they&amp;rsquo;re actually destructive. Something to avoid at all costs.&lt;/p>&lt;p>How can I sustain this? Well, the problem with unit tests is that, by focusing on structure rather than on behaviour, they intrinsically cause your test suite to yield false negatives. So your system works, but your tests fail or don&amp;rsquo;t even compile.&lt;/p>&lt;p>What&amp;rsquo;s the purpose of a table in your kitchen? For you to eat? With how many people? At what time? While watching TV? Can you meet these behavioural needs in a way that employs a different table or table position, or that doesn&amp;rsquo;t require a table at all? Well, if you can, your unit tests will fail. If the height of the table is different, your tests will fail. And if you remove the concept of a table altogether, your tests won&amp;rsquo;t even compile.&lt;/p>&lt;p>So if you change the structure of your code, and you&amp;rsquo;re forced to change your tests, are those tests giving you any confidence at all? How do you know things are still working fine, if you had to change the tests?&lt;/p>&lt;p>Ultimately, unit tests are bad because they prevent you from getting quick and accurate confidence that your changes are good to go, and they increase the amount of work you need to change things.&lt;/p>&lt;p>You might counter with &amp;ldquo;that&amp;rsquo;s kid of okay, no? You structure your service well, and then add coverage to prevent regressions&amp;rdquo;. This is the same as nailing your kitchen table to your kitchen floor, so it stays perfectly still and aligned. But worse, because software is &lt;em>ALL&lt;/em> about changing things. It&amp;rsquo;s never a linear activity, where you build, A, then B, then C, and so on. It&amp;rsquo;s about iterating over and over on the same system, continuously evolving it to react to changing needs.&lt;/p>&lt;h3 id="a-perverse-alternative">A perverse alternative&lt;/h3>&lt;p>Quite a few people and companies have come to realise that unit tests, as described above, are a liability, but the alternative they choose is equally destructive.&lt;/p>&lt;p>They mostly write end-to-end tests, and end up with crazy long test suite runs. Given that end-to-end tests require an environment with the whole system in it, this is typically a shared environment, and things can easily fail because of random IO conditions, or because changes are made to the systems while the tests are still running e.g. when another team deploys a new version of a service.&lt;/p>&lt;p>This solution doesn&amp;rsquo;t yield false positives, but still has false negatives due to its instability, while destroying both the speed and the accuracy of the feedback. If anything, it&amp;rsquo;s worse than the approach that adopts unit tests.&lt;/p>&lt;h2 id="what-to-do-instead">What to do instead&lt;/h2>&lt;p>So approaches that use unit tests or that mostly focus on end-to-end tests are both bad. What are we left with? What can we do instead?&lt;/p>&lt;p>The problems I described derive from a misinterpretation of the concept of unit tests. Yes, they focus on units in isolation, but a &amp;ldquo;unit&amp;rdquo; originally meant a unit of behaviour.&lt;/p>&lt;p>So your unit is not a class or function, but a self-contained module capable of providing value on its own. Not a part of a system, but a sub-system. And these units don&amp;rsquo;t add up linearly, but compose together to form systems of increasingly higher order. Units containing other units, if you will.&lt;/p>&lt;p>What does this mean? It means that an effective test strategy still employs unit, integration, and end-to-end tests, but their responsibilities are separate, and their meaning is different.&lt;/p>&lt;h3 id="end-to-end-tests-1">End-to-end tests&lt;/h3>&lt;p>End-to-end tests are a type of unit tests, where the unit is your whole system.&lt;/p>&lt;p>You want to distinguish between tests that tell you whether your system is currently working, and tests that tell you whether your system works.&lt;/p>&lt;ol>&lt;li>Your system might work, in the sense that it&amp;rsquo;s implemented correctly and handles outages or adverse conditions well.&lt;/li>&lt;li>But it might not currently work, in the sense that a necessary downstream or upstream external system might not be currently working. Or maybe it doesn&amp;rsquo;t have any running instance.&lt;/li>&lt;/ol>&lt;p>It&amp;rsquo;s important that your tests distinguish between these 2, because what you do when your tests fail is different in each case.&lt;/p>&lt;ol>&lt;li>If your system doesn&amp;rsquo;t work correctly, you should change it until it does. This means your pipeline should stop, and you should fix things before releasing the change.&lt;/li>&lt;li>If your system is currently not working, you should look into it operationally, as you might want to get in touch with the companies or teams responsible for your external dependencies. In any case, your build pipelines should not stop, as there&amp;rsquo;s nothing your system is doing wrong.&lt;/li>&lt;/ol>&lt;p>It&amp;rsquo;s useful to adopt different names, and I propose we use &amp;ldquo;smoke tests&amp;rdquo; for tests that tell you whether your system is currently working, and &amp;ldquo;system tests&amp;rdquo; for tests that tell you whether your system works.&lt;/p>&lt;h4 id="smoke-tests">Smoke tests&lt;/h4>&lt;p>You don&amp;rsquo;t typically need many of these, because their function is not to provide confidence that your system works. If any of your upstream or downstream systems isn&amp;rsquo;t currently working, you tend to notice.&lt;/p>&lt;p>They should be deep, to cover your business workflows. An example might involve a new user singing up, then signing in, accessing the value you provide, then logging out, then quitting your product. Another one might excite the whole account recovery workflow. And so on.&lt;/p>&lt;p>You don&amp;rsquo;t even need accuracy here. If anything fails, you can smell smoke, hence the term, and given that your build passed all tests, it&amp;rsquo;s typically a misbehaving upstream or downstream system. You go and investigate. Occasionally you might find that it&amp;rsquo;s not due to an external system, and your tests were not covering something important, so you go and evolve those tests.&lt;/p>&lt;p>Smoke tests are not run as part of your build or even release pipelines, but they run periodically in your production environment. So while you typically don&amp;rsquo;t need many, there&amp;rsquo;s no harm in having more than just a few, as these don&amp;rsquo;t slow down your feedback when you build.&lt;/p>&lt;h4 id="system-tests">System tests&lt;/h4>&lt;p>System tests are end-to-end, similarly to smoke tests, but the similarities end here. These tests tell you whether your system works, so you remove all factors you cannot control, including all upstream or downstream systems that are outside your company.&lt;/p>&lt;p>These are run as part of your build pipeline, in an environment with multiple services in it, and they&amp;rsquo;re slow and cumbersome.&lt;/p>&lt;p>As such, there should be as few of these as possible. The ideal number here is zero. Yeah, you heard that right. If your release pipeline includes error-rate monitoring and canary rollouts with automatic rollbacks, any misconfiguration across services will be caught at that time anyway. But you could still have some, just never rely on these to provide confidence. Use them as an expensive and inconvenient safety net, and get rid of it as soon as you&amp;rsquo;re confident with your test suite.&lt;/p>&lt;h3 id="integration-tests-1">Integration tests&lt;/h3>&lt;p>Integration tests are also a type of unit tests, where the unit is a service, as opposed to your whole system made of multiple services.&lt;/p>&lt;p>If you organize your service to adopt a Hexagonal Architecture - and you should - you can think about your service as some glue, plus a collection of subsystems:&lt;/p>&lt;ol>&lt;li>Driving adapters, which drive your service e.g. HTTP endpoints or Kafka consumers.&lt;/li>&lt;li>Your application, the facade the driving adapters interact with.&lt;/li>&lt;li>Driven adapters, which your application drives e.g. databases, queues, and external APIs.&lt;/li>&lt;/ol>&lt;p>With this in mind, integration tests can be split into &amp;ldquo;adapter tests&amp;rdquo; and &amp;ldquo;service tests&amp;rdquo;.&lt;/p>&lt;h4 id="adapter-tests">Adapter tests&lt;/h4>&lt;p>A port is an external system that either calls your application or is called by it. A driving port or a driven port. An HTTP API or a SQL database.&lt;/p>&lt;p>Your service has an adapter for each of these ports, like an HTTP endpoint, or the JDBC or JPA implementation of a repository.&lt;/p>&lt;p>Adapter tests check that your adapter can adapt your application with their target port. They do not check the correctness of your application.&lt;/p>&lt;h5 id="for-driving-adapters">For driving adapters&lt;/h5>&lt;p>Imagine your application was exposed through an HTTP API, then the set of your HTTP endpoints is your driving adapter for your HTTP API port, or your driving HTTP adapter.&lt;/p>&lt;p>The way you test it is by mocking or stubbing your application to behave in a fixed way, by producing a query or command using a real HTTP client and request, and by then checking that both the request and the response comply with your stated HTTP API e.g. your OpenAPI specification, and that the return data matches what you expected, given the behaviour you fixed in the application and the command or query.&lt;/p>&lt;p>An example might include:&lt;/p>&lt;ol>&lt;li>You want to test your handling of commands to register new users, when a user hasn&amp;rsquo;t already registered with that email address.&lt;/li>&lt;li>You instruct your application layer to return a result stating that the user registration command was accepted successfully (as opposed to another result indicating it was rejected because an already registered user existed).&lt;/li>&lt;li>You then create an HTTP request you can fire with your chosen HTTP client.&lt;/li>&lt;li>You check that your request complies with your OpenAPI specification. Paths, headers, request body schema, etc.&lt;/li>&lt;li>You start your endpoint, injecting your stubbed application in it, and get the port it&amp;rsquo;s running on.&lt;/li>&lt;li>You send the HTTP request to your endpoint, using the HTTP client pointing at the port the endpoint is running on.&lt;/li>&lt;li>You receive an HTTP response, and check it complies with your OpenAPI specification as a response for the given request. Paths, headers, status code, response body schema, etc.&lt;/li>&lt;li>You check that the data in the response matches in content what you expect, based on the use case e.g. that it indicates that things were accepted just fine. This involves the expected status code e.g. 202, and might also involve the headers and response body.&lt;/li>&lt;/ol>&lt;p>These tests still run locally and are quite fast (~1 second), and you typically have 1 suite per adapter, requiring to start each driving adapter e.g. the HTTP server only once.&lt;/p>&lt;h5 id="for-driven-adapters">For driven adapters&lt;/h5>&lt;p>Imagine your application wrote to a SQL database or appended an event to an event bus. The database or event bus is your port, and your driven adapter is the logic that transforms a call in the domain language your application speaks into a call to this external service.&lt;/p>&lt;p>The way you test it is by creating preconditions in your downstream system, by driving an invocation using your domain-level facade e.g. a repository interface, and by then checking the result returned by your domain-level facade and then internal state of your downstream dependency.&lt;/p>&lt;p>An example might include:&lt;/p>&lt;ol>&lt;li>You want to test recording a newly registered user, either by inserting it in a Postgres database, or by appending an event to Kafka.&lt;/li>&lt;li>You start Postgres or Kafka in Docker using &lt;code>TestContainers&lt;/code>, and use a raw JDBC client or Kafka producer to create preconditions in the form of preexisting data.&lt;/li>&lt;li>You create an instance of your &lt;code>PostgresUserRepository&lt;/code> or &lt;code>KafkaUserEventPublisher&lt;/code>, pointing at your running Postgres or Kafka container.&lt;/li>&lt;li>You use your &lt;code>UserRepository&lt;/code> domain-level interface to record the fact that a user was registered.&lt;/li>&lt;li>You check that the value returned by the domain-level interface matches your expectations, given the preconditions.&lt;/li>&lt;li>You use your raw JDBC client or Kafka client to check the state of your Postgres or Kafka container after the invocation e.g. that the data is there, or that you can receive the event on the right topic.&lt;/li>&lt;/ol>&lt;p>These tests still run locally and are reasonably fast (~1-5 seconds), and you typically have 1 suite per adapter, requiring you to start each driven adapter e.g. Postgres or Kafka in Docker only once.&lt;/p>&lt;h4 id="service-tests">Service tests&lt;/h4>&lt;p>Service tests test your service as a whole, mostly covering the glue code and your service-level configuration.&lt;/p>&lt;p>Typically, a service test:&lt;/p>&lt;ol>&lt;li>Starts the driven ports e.g. your Postgres database or your Kafka cluster in Docker, using &lt;code>TestContainers&lt;/code>.&lt;/li>&lt;li>Creates some preexisting data inside these ports using raw clients.&lt;/li>&lt;li>Starts the whole service, like it would be started externally, by passing args and environment variables to the main class or function.&lt;/li>&lt;li>Uses a raw client to drive the service e.g. an HTTP client.&lt;/li>&lt;li>Checks that the response has the right status and data.&lt;/li>&lt;/ol>&lt;p>These tests still run locally and are reasonably fast (~1-5 seconds), and you typically have 1 suite for a whole service. These don&amp;rsquo;t duplicate the scenarios tested elsewhere, because their job is not to catch poorly implemented domain or integration logic, but to catch problems with the glue code or the configuration. This means there&amp;rsquo;s usually 1-2 cases here, the main happy path and handling unavailable dependencies at service start time.&lt;/p>&lt;h5 id="in-process-and-in-docker">In-process and in-Docker&lt;/h5>&lt;p>Ideally you want to run the service tests both in-process and in-Docker.&lt;/p>&lt;ul>&lt;li>The in-process tests start the service within the same process the test runs in e.g. by calling your main function if you use Kotlin or Java.&lt;/li>&lt;li>The in-Docker tests start the service within a Docker container, as a separate process. This requires building the Docker image as part of your local build, and then using &lt;code>TestContainers&lt;/code> to start the current version of your service within a Docker container.&lt;/li>&lt;/ul>&lt;p>Ideally you should use the same test specification for both, and the only difference should be how you start the service.&lt;/p>&lt;p>Also testing in-Docker allows you to catch problems related to how you build, configure, and run your Docker container. This is important, because services tend to run packaged these days, so you better catch packaging issues early.&lt;/p>&lt;h3 id="unit-tests-1">Unit tests&lt;/h3>&lt;p>Unit tests focus on testing your application and domain logic. Instead of focusing on whether a given type works, a unit test covers a unit of behaviour, i.e. a subsystem or module within your application or domain layers.&lt;/p>&lt;p>This requires structuring your application into modules that are self-contained, each with their own contract. You should treat these modules like libraries.&lt;/p>&lt;p>Each module get its own tests, checking that the module behaves correctly against its contract. What contract am I talking about? Here there&amp;rsquo;s no OpenAPI or Avro after all? I&amp;rsquo;m talking about its public API.&lt;/p>&lt;p>A module will have a contract or API, which is the set of all its publicly visible types and functions.&lt;/p>&lt;h4 id="contract-tests">Contract tests&lt;/h4>&lt;p>Contract tests are a type of tests that test a module against its contract, or public API. These are basically runnable specifications, documentation, and examples of how the module delivers its value through the interaction of its types and functions.&lt;/p>&lt;p>Every workflow or condition is represented in these tests, but only publicly visible types and functions can appear here.&lt;/p>&lt;p>These tests run locally, in-memory, and have no IO interactions, meaning they&amp;rsquo;re really fast and reliable.&lt;/p>&lt;p>When you build a library, these example tests are also hugely useful for the users of your library.&lt;/p>&lt;h4 id="type-specific-tests">Type-specific tests&lt;/h4>&lt;p>Every publicly visible type might also get its type-specific test, to check that type behaves correctly. This is not as important in a service as it is in a library.&lt;/p>&lt;p>In a library, each publicly visible type gets its own test, because you don&amp;rsquo;t know how what the users of the library will do. In a service, contract tests are typically enough, because if a type has an issue but the contract tests pass, it means that the issue is not impacting your expected behaviour.&lt;/p>&lt;h2 id="final-words">Final words&lt;/h2>&lt;p>That&amp;rsquo;s it, a long explanation of my go-to test strategy, and of why I believe it&amp;rsquo;s a good idea compared to some popular approaches. I apologise to Kent Beck for using the word &amp;ldquo;unit tests&amp;rdquo; with a negative connotation, but sadly these days they tend to mean something far from their original meaning.&lt;/p>&lt;p>I leave you with some further advice you might find useful when structuring your test strategy:&lt;/p>&lt;ol>&lt;li>Don&amp;rsquo;t aim for coverage, but at getting quick and accurate confidence that your changes are good to go.&lt;/li>&lt;li>Ensure you can run most of your tests locally.&lt;/li>&lt;li>Ensure your tests can survive structural changes without false negatives or broken compilation.&lt;/li>&lt;li>Ensure each test type is specific, removing the need to cover possibilities in a cumbersome and slow way, when they could have been covered in a faster and more convenient way instead.&lt;/li>&lt;li>Write your tests first, and your application and integration logic second, in small steps, using Test-Driven Development.&lt;/li>&lt;li>Organize and package your services and systems for ease of testing things without resorting to end-to-end system tests. It will pay off big time.&lt;/li>&lt;/ol></description></item><item><title>Developer productivity: what's all the fuss about?</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-31-developer-productivity-fuss/</link><pubDate>Thu, 31 Aug 2023 09:30:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-31-developer-productivity-fuss/</guid><description>&lt;h2 id="the-controversy-about-developer-productivity">The controversy about developer productivity&lt;/h2>&lt;p>There&amp;rsquo;s been a lot of controversy recently about developer productivity. Some very big consultancies claimed you could and should measure it, while some software product development legends are of a different opinion.The legends are making fun of the consultancies, which are in turn accusing the legends of wanting to dodge accountability for their work.&lt;/p>&lt;p>So who&amp;rsquo;s right? Who&amp;rsquo;s wrong? And why? To answer that we&amp;rsquo;ll have to think a bit deeper.&lt;/p>&lt;h2 id="a-framework-to-think-about-developer-productivity">A framework to think about developer productivity&lt;/h2>&lt;p>In order to reason about developer productivity, we should first ask ourselves a few questions.&lt;/p>&lt;ol>&lt;li>What is developer productivity? What&amp;rsquo;s its operational definition?&lt;/li>&lt;li>Is the productivity of a software developer intrinsic to the person? Or does it depend mostly on other factors?&lt;/li>&lt;li>Assuming you could measure the productivity of a software product developer, would this be useful? To who? How?&lt;/li>&lt;/ol>&lt;p>As always, we start with why.&lt;/p>&lt;h3 id="imagine-you-could-measure-it">Imagine you could measure it&lt;/h3>&lt;p>Let&amp;rsquo;s assume you could come up with an accurate productivity score for a developer, on a comforting 0-100 scale. Now, what would you do with this information? Well, it really depends on what your role within the company is.&lt;/p>&lt;ul>&lt;li>If you&amp;rsquo;re a developer, and somehow knew this score, you could use it as an indicator, to experiment with different things and see what improves it.&lt;/li>&lt;li>As a manager, you could use this score to understand the consequences of your actions, and determine who needs coaching and attention.&lt;/li>&lt;li>As a CTO or head of department, you could keep an eye on the average score across the teams, to get feedback on your hiring policies and training efforts.&lt;/li>&lt;/ul>&lt;p>There&amp;rsquo;s nothing intrinsically wrong with metrics, but they can easily be abused.&lt;/p>&lt;ul>&lt;li>As a developer, you could use your higher score to feel entitled to special treatment. To expect that your opinion carried more weight, compared to the ideas of others with a lower score.&lt;/li>&lt;li>If you&amp;rsquo;re a manager, you could use this score blindly to promote and fire, without looking into why it might differ across people, and without asking yourself whether the person might enjoy or even do well in the new role.&lt;/li>&lt;li>As a CTO, you could use this score to second guess your direct reports about their own employees&amp;rsquo; performance. Or you could assign a target average score as a goal to a manager for their area or team.&lt;/li>&lt;/ul>&lt;h3 id="is-the-productivity-of-a-developer-an-intrinsic-trait">Is the productivity of a developer an intrinsic trait?&lt;/h3>&lt;p>Let&amp;rsquo;s assume you accurately measure developer productivity on a 0-100 scale. Now, what would this depend on? Some people immediately associate the productivity of an employee to the merits or defects of that employee. Like if it was an intrinsic trait, something they come with.&lt;/p>&lt;p>But does productivity really belong to the employee? Or is it largely influenced by the environment: by the team, the ways of working, the salary and benefits, the technologies, the state of the codebase, the processes, the manager, and the company?&lt;/p>&lt;p>It should hopefully be obvious that these factors largely influence the productivity of a developer. So if the same person changes team, company, processes, manager, or ways of working, their productivity will also likely change, sometimes dramatically.&lt;/p>&lt;h3 id="developer-productivity-what-does-it-even-mean">Developer productivity: what does it even mean?&lt;/h3>&lt;p>Why are we even talking about individual developer productivity anyway? After all software product development it&amp;rsquo;s collaborative knowledge work.&lt;/p>&lt;p>First of all, developers are service providers and knowledge workers. They do not produce anything. Their effectiveness is never measurable as stuff over time. Sounds crazy? Are you thinking that, as a developer myself, I&amp;rsquo;m simply trying to avoid accountability?&lt;/p>&lt;p>Software developers are not the only kind of knowledge workers, so this shouldn&amp;rsquo;t feel like a new concept. What about lawyers, judges, scientists, city planners, advisors, doctors, teachers, and psychologists? How do you define or measure their productivity?&lt;/p>&lt;p>About scientists, would you go with number of publications a year? Number of whiteboards filled? How about judges? How quickly they judge somebody?&lt;/p>&lt;p>There&amp;rsquo;s this idea that each employee produces an amount of stuff over a unit of time, that stuff adds up to outcomes and impact, and that you can thus infer how much an individual contributes to the baseline based on how much stuff they produce over time. Well, in software product development this is ridiculous and absurd.&lt;/p>&lt;p>Also, software development is a collaborative job. In a collaborative activity, the determining factor for performance is how people interact, rather than how they act individually.&lt;/p>&lt;p>Think about a team of mechanics that services a Formula 1 car at a pit stop: their alignment and coordination trumps how well they perform in isolation. A well-coordinated team of less skilled mechanics would easily be more effective than the other way around.What about a football team? Or an army squad? Well, you should think about software product developers the same way.&lt;/p>&lt;p>And you know what else is true about every collaborative activity? That effectiveness is improved by teaming together people with different strengths. Teams made of specialists that excel at different things routinely outperform homogeneous teams of all-rounders.So you cannot even compare the performance of people, because everybody is different. There&amp;rsquo;s not merely 1 dimension to this problem, but many. Even within the same role and area of strength, the same person can work out wildly differently across 2 different teams, because of the strength of the interactions with the other team members.&lt;/p>&lt;h2 id="what-should-we-do-about-performance-then">What should we do about performance then?&lt;/h2>&lt;p>This controversy is the result of a different world view. On one side you have Theory X people, with their internal competition model, claiming you cannot manage what you cannot measure. On the opposite side, you have Theory Y people, talking about teams, systems, and culture.&lt;/p>&lt;p>For managers who subscribe to Theory X, the only explanation for any undesired condition is that some of their employees must be incompetent or a slacker. It&amp;rsquo;s always somebody else. They see their job as identifying, removing, and replacing people. &amp;ldquo;If each of the 5 team members produces 3, I have a productivity of 15, so if I fired a 3 and hired a 4 I&amp;rsquo;d bring it to 16 and improve productivity. These damn 3s are holding me down!&amp;rdquo;&lt;/p>&lt;p>Theory Y isn&amp;rsquo;t that comforting.&lt;/p>&lt;p>As a Theory Y manager, the only person you can blame for any undesired condition with your team is yourself. &amp;ldquo;They&amp;rsquo;re not giving it all, I failed to motivate them&amp;rdquo;. &amp;ldquo;They&amp;rsquo;re not competent, I failed at both hiring and training&amp;rdquo;. &amp;ldquo;We got a lot of bugs and rework, I failed at structuring effective processes and at giving the right incentives&amp;rdquo;.&lt;/p>&lt;p>Theory Y puts performance issues squarely in the management&amp;rsquo;s lap. Like a professional sports coach, you might occasionally replace somebody, but your main focus should be directed at improving performance, rather than obsessing over measurements. Even if people decide to measure something, the metrics should be indicators, never targets. What matters the most is continuous improvement, fostered by vigorous training, to ensure that you work effectively together.&lt;/p>&lt;p>Unfortunately subscribing to Theory Y is very hard for a manager, because it shifts the accountability Theory X people love talking about from the employees to themselves. And if your job is about improving the effectiveness of your team, rather than measuring, blaming, and firing, then it follows that, in order to do your job, you should know the work intimately.&lt;/p>&lt;p>And these 2 points, ladies and gentlemen, are at the core of this controversy, which has nothing to do with developer productivity, and everything to do with Theory X and Theory Y. And that&amp;rsquo;s why those &amp;ldquo;managers&amp;rdquo; that know nothing about the work of the people they&amp;rsquo;re supposed to manage cling so much to measurements, scores, and KPIs: because what else do they have? It&amp;rsquo;d become evident that they&amp;rsquo;re good for nothing, should Theory Y management practices ever be introduced in their companies.&lt;/p></description></item><item><title>Array sorters aren't problem solvers</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-27-array-sorters-arent-problem-solvers/</link><pubDate>Sun, 27 Aug 2023 21:30:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-27-array-sorters-arent-problem-solvers/</guid><description>&lt;h2 id="the-evergreen-problem-solving">The evergreen problem-solving&lt;/h2>&lt;p>If you work in software product development like me, you hear a lot about problem-solving. Problem-solving skills, problem-solving attitude, problem-solving interview steps.&lt;/p>&lt;p>What most companies call problem-solving has nothing to do with actual problems, though. And the interview steps aimed at testing a candidate&amp;rsquo;s skills at this? They&amp;rsquo;re completely off the mark. As a candidate, you typically get 30 to 45 minutes to complete a coding task, which usually involves sorting arrays, inverting binary trees, or achieving something trivial under some made-up limitations, like implementing the multiplication operation.&lt;/p>&lt;p>So what&amp;rsquo;s wrong with these questions? Are these not problems? Not at all, they&amp;rsquo;re exercises.&lt;/p>&lt;h2 id="problems-vs-exercises">Problems vs exercises&lt;/h2>&lt;p>An exercise is always devoid of context, static, and timeless. It&amp;rsquo;s something you can analyse and work on in isolation, and it&amp;rsquo;s immediately obvious whether a potential solution works for an exercise.&lt;/p>&lt;p>Problems are a different story. An example of a problem might be &amp;ldquo;why do we keep producing so many bugs, irrespective of all our efforts and of how much the CEO cares about this?&amp;rdquo; Problems are messy, and always come with a lot of context, ambiguity, and related problems. It&amp;rsquo;s not clear whether a candidate solution will work, or whether a solution exists at all.&lt;/p>&lt;p>Successful problem-solving requires finding the right solution to the right problem. We fail more often because we solve the wrong problem than because we get the wrong solution to the right problem. Our ability to solve problem is limited by our conception of what is feasible.&lt;/p>&lt;p>Analysis works well for exercises:&lt;/p>&lt;ol>&lt;li>You break down the exercise into its constituting parts.&lt;/li>&lt;li>You recursively analyse the parts.&lt;/li>&lt;li>You find how to make each part of the exercise work.&lt;/li>&lt;li>You put the parts together and, voilà, your exercise is solved.&lt;/li>&lt;/ol>&lt;p>Synthesis is much better suited for problems:&lt;/p>&lt;ol>&lt;li>You look at the containing system the problem manifested itself in.&lt;/li>&lt;li>You recursively look at the containing systems e.g. the team, the department, the company, the market, the city, the country, the industry.&lt;/li>&lt;li>You understand why the problem manifested itself in the existing system.&lt;/li>&lt;li>You redesign the system so that the problem cannot manifest itself: you dissolve it.&lt;/li>&lt;/ol>&lt;p>Redesigning and changing a system are problems themselves, so you never know what will work. Two companies might require a very different approach to solve the &amp;ldquo;same&amp;rdquo; problem.&lt;/p>&lt;h2 id="the-state-of-the-industry">The state of the industry&lt;/h2>&lt;p>So what&amp;rsquo;s the big deal? Well, it&amp;rsquo;s that array sorters are not problem solvers. And being able to sort an array in 20 minutes, or implementing Timsort by hand on a whiteboard, says absolutely nothing about your ability to solve problems. If anything, people who obsess over exercises tend not to do so well with problems.&lt;/p>&lt;p>No company ever went bust because their development team couldn&amp;rsquo;t sort an array. No customer refused to renew their subscription to a company&amp;rsquo;s product because people couldn&amp;rsquo;t invert a binary tree. In comparison, plenty of companies died because they could not solve the social dynamics between their founders, or because they allowed entropy to grow too much and ended up having to run just to stand still.&lt;/p>&lt;p>Why do companies insist with exercises then? Because in a world that loves pre-canned, microwavable solutions, understanding and wisdom are often dismissed as wanting to overcomplicate things. Because you cannot appreciate things you don&amp;rsquo;t understand, and understanding and wisdom are an exception.&lt;/p>&lt;p>And because, if you can&amp;rsquo;t design a real-scale Twitter in 40m, you clearly won&amp;rsquo;t be able to move JSON objects from the company&amp;rsquo;s web app to the company&amp;rsquo;s database and back.&lt;/p></description></item><item><title>Confessions of a TDD fanboy</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-06-tdd-confessions/</link><pubDate>Sun, 06 Aug 2023 16:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-08-06-tdd-confessions/</guid><description>&lt;p>Test-Driven Development (TDD) is arguably one of the most polarizing and misunderstood approaches in software development. It&amp;rsquo;s love it or hate it.&lt;/p>&lt;p>Let&amp;rsquo;s have a look at TDD and why it provokes such strong reactions. This is going to piss some people off&amp;hellip;&lt;/p>&lt;h1 id="so-what-is-test-driven-development">So what is Test-Driven Development?&lt;/h1>&lt;p>Put down in words by &lt;a href="https://en.wikipedia.org/wiki/Kent_Beck">Kent Beck&lt;/a> in 2003, Test-Driven Development is a software development process relying on tests to drive the implementation of the code.&lt;/p>&lt;p>The approach adopts the following sequence to introduce changes in behaviour:&lt;/p>&lt;h2 id="1-add-a-new-test">1. Add a new test&lt;/h2>&lt;p>Start by writing a new test case.&lt;/p>&lt;ul>&lt;li>1 test case, not a whole suite, not all requirements upfront.&lt;/li>&lt;li>These tests are not unit tests about a specific type. There are collaborative tests, where different types and functions interact to achieve a result. A bad name would be &lt;code>ShoppingCartTest&lt;/code>, while a good name would be &lt;code>ShoppingExperienceTests&lt;/code>.&lt;/li>&lt;li>If you&amp;rsquo;re starting a project or a new module, nothing will even compile at first, and that&amp;rsquo;s fine. Write things as you&amp;rsquo;d want them, and then make it compile.&lt;/li>&lt;li>Not talking about BDD here. Tests should be expressed in your main programming language, without a DSL.&lt;/li>&lt;li>Start from the assertion, thinking about what needs to happen. Try to design the test so that there&amp;rsquo;s a single assertion.&lt;/li>&lt;li>It&amp;rsquo;s important to write the test so that it provides an example of how the code should be used in that case. No cryptic test functions to template test cases.&lt;/li>&lt;/ul>&lt;h2 id="2-run-all-tests-the-new-test-should-fail-for-the-expected-reasons">2. Run all tests. The new test should fail for the expected reasons&lt;/h2>&lt;p>Then run all the tests, and confirm the new test case is failing for the expected reasons.&lt;/p>&lt;ul>&lt;li>This should be fast.&lt;ul>&lt;li>Minimize the number of tests that interact with IO.&lt;/li>&lt;li>A smart build system e.g. Gradle can greatly help by only running tests for the modules that actually changed.&lt;/li>&lt;li>Adopt a modular component-level software architecture e.g. Hexagonal Architecture to achieve crisp module boundaries.&lt;/li>&lt;/ul>&lt;/li>&lt;li>Don&amp;rsquo;t skip the part where you make the new test fail: it prevents you from assuming things work in presence of a mistake in the test logic or in the implementation.&lt;/li>&lt;/ul>&lt;h2 id="3-write-the-simplest-code-that-passes-the-new-test">3. Write the simplest code that passes the new test&lt;/h2>&lt;p>Then write an implementation that passes the new test.&lt;/p>&lt;ul>&lt;li>Don&amp;rsquo;t worry too much about code structure at this stage.&lt;/li>&lt;li>Once the test passes, changing the structure will be safer and faster.&lt;/li>&lt;li>Don&amp;rsquo;t get obsessed with the word &amp;ldquo;simplest&amp;rdquo; here.&lt;ul>&lt;li>Some folks would return stubbed data to make the test pass, just to then change it.&lt;/li>&lt;li>Apply any good design patterns at this stage as well (don&amp;rsquo;t make things mutable, etc.).&lt;/li>&lt;li>Just don&amp;rsquo;t focus much on finding the right abstractions for the internal code.&lt;/li>&lt;/ul>&lt;/li>&lt;/ul>&lt;h2 id="4-all-tests-should-now-pass">4. All tests should now pass&lt;/h2>&lt;p>Run all tests again, and confirm you haven&amp;rsquo;t caused any other test to fail. If you did, back to #3.&lt;/p>&lt;ul>&lt;li>Once again, this should be fast.&lt;/li>&lt;li>Time to commit after all tests pass.&lt;/li>&lt;/ul>&lt;h2 id="5-refactor-as-needed-using-tests-after-each-refactoring-to-ensure-that-functionality-is-preserved">5. Refactor as needed, using tests after each refactoring to ensure that functionality is preserved&lt;/h2>&lt;p>Refactoring means changing the structure of the code without changing its behaviour. Improve the structure of the implementation, running all tests every time to confirm you haven&amp;rsquo;t changed the behaviour.&lt;/p>&lt;ul>&lt;li>Use small steps here.&lt;/li>&lt;li>Run all tests after each step, not only the test you&amp;rsquo;re using to drive the implementation.&lt;/li>&lt;li>Commit after each step.&lt;/li>&lt;/ul>&lt;h1 id="so-whats-to-love">So what&amp;rsquo;s to love?&lt;/h1>&lt;p>So what&amp;rsquo;s the big deal with Test-Driven Development? Why might this be a good idea?&lt;/p>&lt;p>Many people focus on the tests, claiming you could always write them later, but TDD isn&amp;rsquo;t about testing. TDD is a software development process: its goal is to drive code design and facilitate the implementation. The tests you write are a desirable side effect, not the goal.&lt;/p>&lt;h2 id="backward-design">Backward design&lt;/h2>&lt;p>By writing a new test case first, you focus more on the requirement or desired results. This helps because your work is more intentional, rather than building things and figuring out why later.&lt;/p>&lt;p>Also, you write the code like you&amp;rsquo;d want to use it, to make your own life easy while writing the test, and to create a readable example of how the code would be used to achieve something. This is a tremendous difference, as how you want the code to behave drives your implementation choices: types, names, functions, arguments, etc. Without this is hard to come up with good design upfront, as you don&amp;rsquo;t visualize the result and the alternatives.&lt;/p>&lt;h2 id="many-more-smaller-steps">Many more smaller steps&lt;/h2>&lt;p>By cycling through the steps quickly and in small steps, you minimize risk. The fact that you commit often, and that you increment the expected behaviour 1 test at a time means that undoing something is always quick and painless.&lt;/p>&lt;p>Overall it&amp;rsquo;s just harder to get things wrong, and TDD offers frequent rewards, when all tests are green and you commit. It offers a sense of progress and achievement.&lt;/p>&lt;p>It&amp;rsquo;s kind of fun really, far from the hair-pulling debugging some people are used to.&lt;/p>&lt;h2 id="its-fast">It&amp;rsquo;s fast!&lt;/h2>&lt;p>This is very surprising for most people, and a counter-intuitive fact: Test-Driven Development is fast! The focus on the requirements, structuring the code well, decomposing the problem, and having tests to help you find issues earlier, all ultimately result in increased speed.&lt;/p>&lt;p>Producing a correct implementation is quicker, not to mention how quicker it is compared to producing a correct implementation AND writing the tests for it.&lt;/p>&lt;h2 id="the-code-ends-up-of-higher-quality">The code ends up of higher quality&lt;/h2>&lt;p>TDD won&amp;rsquo;t turn you into a great software designer if you&amp;rsquo;re not, but it&amp;rsquo;ll likely bring out the best in you in this sense.&lt;/p>&lt;p>Writing the test case upfront like an example of how the code should be used:&lt;/p>&lt;ul>&lt;li>Forces you to think about its usage.&lt;/li>&lt;li>Forces you to write code that needs to work elegantly in at least 2 places (in the tests, and in the main application).&lt;/li>&lt;li>Produces crisp abstractions, as they are written before any thinking about the implementation, and thus are much less likely to end up leaking implementation details.&lt;/li>&lt;li>Forces you to write testable code. This is not only good for testing. Code is easy to test when it&amp;rsquo;s decoupled and well-modularized. So if you&amp;rsquo;re forced to write testable code, you get low coupling and effective modules.&lt;/li>&lt;li>Results in a stronger and more effective domain model. This is a result of aiming for readable code, and of having to make your life easier by reducing the amount of things to cover, by leveraging the compiler and introducing self-validating types. Separating API and implementation is also easier, as you design these during different steps of the TDD cycle.&lt;/li>&lt;/ul>&lt;h2 id="it-produces-good-tests-as-a-side-effect">It produces good tests as a side effect&lt;/h2>&lt;p>Yeah, let&amp;rsquo;s face it, TDD is not about testing but the tests it produces are a very nice side effect.&lt;/p>&lt;p>The tests produced are great, because:&lt;/p>&lt;ul>&lt;li>They&amp;rsquo;re executable examples, acting as both documentation and regression tests. As they&amp;rsquo;re written upfront, they tend to be readable and self-explanatory.&lt;/li>&lt;li>They are black-box, focusing on contracts and behaviour, rather than on structure and specific types. The result is that you can change the internal structure without your tests getting broken in the process. A far cry from unit tests here.&lt;/li>&lt;li>They offer a high degree of confidence. By adding behaviour incrementally you progressively improve the confidence in your tests, catching problems early, and refining their effectiveness.&lt;/li>&lt;/ul>&lt;h1 id="so-whats-to-hate">So what&amp;rsquo;s to hate?&lt;/h1>&lt;p>For all the reasons above, I firmly believe that Test-Driven Development is a great approach to producing high quality software that works, quickly, while having fun. Mind that I&amp;rsquo;m not a consultant. I write code as a part of my job, and I find TDD invaluable. It brings out the best in me, and helps me get better and better. I&amp;rsquo;m gutted that I learned TDD late, and I had to do this &amp;ldquo;by myself&amp;rdquo;, through books, courses, and videos.&lt;/p>&lt;p>And yet many people are very vocal about their hate and distrust for Test-Driven Development? Why? Well, I wouldn&amp;rsquo;t really know to be honest. But I can guess.&lt;/p>&lt;p>There&amp;rsquo;s an element of wanting to belong to a community, so some folks attack things to feel like they belong with other like-minded folks. Some others had bad experiences from meeting fanatics or people who claimed to do TDD but showed something else.&lt;/p>&lt;p>Some other people, simply put, don&amp;rsquo;t value code structure at all. They think their job is about writing code that works, and get satisfaction from being &amp;ldquo;clever&amp;rdquo; with algorithms and approaches.&lt;/p>&lt;p>The reality is that in software product development code quality matters more than code correctness. I&amp;rsquo;d take high-quality incorrect code over low-quality correct and performant code any day.&lt;/p>&lt;p>Why? Because the job isn&amp;rsquo;t about making things work. Except for very few companies, making things work is trivial. Others have done it already. There are books, blog posts, videos, examples, and tutorials.&lt;/p>&lt;p>In most contexts, implementing a sorting algorithm or inverting a linked list manually should be a fireable offense. Your job is to develop a product that does the job well for your customers, under their circumstances. The less code involved, the better.&lt;/p>&lt;p>And yet some folks buy this myth of the smart lone wolf developer, and pride themselves in their ability to endure long hours of debugging convoluted and obscure logic that solves hard problems. It&amp;rsquo;s people who spend time on LeetCode, or who post unreadable crappy code, asking other people to explain what the console prints. It&amp;rsquo;s people who ask candidates in interviews to manually sort things, or to design Twitter in 1 hour.&lt;/p>&lt;p>Except for the fact that LeetCode &amp;ldquo;problems&amp;rdquo; are not problems, but exercises. Exercises are always environment-free. You have to invert a binary tree, or sort something. When there are constrains, they&amp;rsquo;re made up, like you cannot use some functions from the standard library.&lt;/p>&lt;p>Problems, on the other hand, never come isolated, they have interdependencies with other problems, and they&amp;rsquo;re very context-specific. No company ever went out of business because they couldn&amp;rsquo;t solve an exercise.&lt;/p>&lt;p>The difficult part of software product development is figuring out how to deliver value to users while not imploding from increasing entropy. It&amp;rsquo;s keeping the codebase understandable and easy to change. It&amp;rsquo;s not taking 3 steps forward and 2 steps back. It&amp;rsquo;s being able to contribute to a codebase with 50 developers working in teams. It&amp;rsquo;s not ending up taking a month for something that used to take 1 week the year before.&lt;/p>&lt;p>The best thing about Test-Driven Development is that it removes the drama and the heroics from this profession. It&amp;rsquo;s a methodical approach that removes the need to get things right the first time, producing the ability to safely and quickly iterate without regressions until things work and are structured effectively.&lt;/p>&lt;p>And that&amp;rsquo;s what the LeetCode fanboys can&amp;rsquo;t stomach. It&amp;rsquo;s like pulling the rug from under their feet.&lt;/p>&lt;p>&amp;ldquo;How dare you claim that my hand-made red-black tree implementation isn&amp;rsquo;t going to be a competitive advantage for my company&amp;rdquo;?&lt;/p>&lt;p>&amp;ldquo;How dare you claim that my untested, untestable, cryptic, hard to change minefield of an implementation could ever not be perfect&amp;rdquo;?&lt;/p>&lt;p>And there you have it folks, some people hate TDD because they don&amp;rsquo;t understand what the job is about. They don&amp;rsquo;t consider it a good solution, because they&amp;rsquo;re not aware of the problem.&lt;/p></description></item><item><title>To modularize or not to modularize, that is the question!</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-21-to-modularise-or-not-to-modularise/</link><pubDate>Fri, 21 Jul 2023 17:30:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-21-to-modularise-or-not-to-modularise/</guid><description>&lt;p>&lt;em>Why are we talking about modules?&lt;/em>&lt;/p>&lt;p>Because we want to scale, and if we don&amp;rsquo;t modularize we won&amp;rsquo;t be able to scale.&lt;/p>&lt;p>&lt;em>Mmm, but wait a minute, there are giant companies that don&amp;rsquo;t use modules, so we can scale without them!&lt;/em>&lt;/p>&lt;p>Scaling doesn&amp;rsquo;t mean becoming bigger and bigger, kind of the opposite of that.&lt;/p>&lt;p>&lt;em>What about microservices though? Aren&amp;rsquo;t those kind of modules? Don&amp;rsquo;t they allow to re-use code?&lt;/em>&lt;/p>&lt;p>Nope. They don&amp;rsquo;t. That&amp;rsquo;s why very few companies can re-use the capabilities they develop and achieve scaling.&lt;/p>&lt;h1 id="what-scaling-means">What “scaling” means&lt;/h1>&lt;h2 id="not-becoming-bigger">Not becoming bigger&lt;/h2>&lt;p>&lt;em>So what does “scaling” means? How does it not mean becoming bigger?!&lt;/em>&lt;/p>&lt;p>Scaling means achieving more and more outcomes, without increasing costs and efforts as much.&lt;/p>&lt;p>&lt;em>Mmm, can you elaborate?&lt;/em>&lt;/p>&lt;p>Of course! Imagine you create a new company, you launch some products, something sticks, and you make some profits.&lt;/p>&lt;p>&lt;em>OK, so what?&lt;/em>&lt;/p>&lt;p>You might be tempted to say, hey, if we have 5 teams that make 2 products, and that makes us £1M a year in profit, if we get to 10 teams and 4 products, then we&amp;rsquo;ll make £2M a year in profit. We need to scale!&lt;/p>&lt;p>&lt;em>Isn&amp;rsquo;t that true?&lt;/em>&lt;/p>&lt;p>Well, at best that would be a very inefficient way of growing, but in practice it never works.&lt;/p>&lt;p>&lt;em>Why not?&lt;/em>&lt;/p>&lt;p>It&amp;rsquo;s a long story. As an example, the new products you launch might not be as profitable. Or the new features you add to your existing products might not improve your profitability or market share, or at least not as much as it&amp;rsquo;d take to justify your efforts to add these.&lt;/p>&lt;p>&lt;em>But we have a great product function, surely we can tackle these issues.&lt;/em>&lt;/p>&lt;p>Perhaps, yeah, but there&amp;rsquo;s a more fundamental problem. Complexity.&lt;/p>&lt;p>&lt;em>Complexity?&lt;/em>&lt;/p>&lt;p>Yeah. Basically the effort required to run a company or a system increases super-linearly with its complexity. And size is a good proxy for complexity, for both companies andsystems.&lt;/p>&lt;p>&lt;em>What are you saying?&lt;/em>&lt;/p>&lt;p>That the more people you have, the more effort you need to run a company, and this grows faster than the number of people, so your capacity shrinks.&lt;/p>&lt;p>&lt;em>Mmm, capacity? What do you mean?&lt;/em>&lt;/p>&lt;p>The people you hire in a company generate capacity, and you use capacity to produce outputs. These outputs, if you&amp;rsquo;re lucky, result in outcomes.&lt;/p>&lt;p>&lt;em>Ah, gotcha, I get what capacity is, it&amp;rsquo;s our ability to do stuff. But what do you mean it shrinks as we add people.&lt;/em>&lt;/p>&lt;p>It does, unfortunately. A company of 100 requires more effort to run than 5 times the effort a 20-people company needs.&lt;/p>&lt;p>&lt;em>Really? Why?&lt;/em>&lt;/p>&lt;p>Many reasons. Management hierarchies become longer, communication slows down, nobody has the full context anymore, you need more processes, more tools, more controls, etc. So the capacity you get from people diminishes (because of diluted culture, processes, controls,etc.), you spend more of that capacity to run the company (implementing controls, running tools, managing people, etc.), and your outputs don&amp;rsquo;t translate to outcomes as much (looser alignment, features nobody wanted, etc.).&lt;/p>&lt;p>&lt;em>Okay, makes sense. I actually think we suffer from this now! But what&amp;rsquo;s the alternative really?&lt;/em>&lt;/p>&lt;h2 id="but-achieving-more-with-less">But achieving more with less&lt;/h2>&lt;p>The whole idea of scaling is about achieving more without increasing capacity as much.&lt;/p>&lt;p>&lt;em>So are you saying we could achieve more with the same capacity, meaning the same number of people?&lt;/em>&lt;/p>&lt;p>Yes! And every company needs to achieve this by the way!&lt;/p>&lt;p>&lt;em>So we won&amp;rsquo;t increase our capacity?&lt;/em>&lt;/p>&lt;p>Oh no, we will increase it, but by working better, rather than by hiring more.&lt;/p>&lt;p>&lt;em>But will that be enough?&lt;/em>&lt;/p>&lt;p>It might, or it might not, but capacity is not the whole picture. How you use capacity matters a lot.&lt;/p>&lt;p>&lt;em>What do you mean?&lt;/em>&lt;/p>&lt;p>Capacity is how much effort you can exert, but outcomes depends on how much effort outputs take, and how much these outputs contribute to outcomes.&lt;/p>&lt;p>&lt;em>Efficiency and effectiveness?&lt;/em>&lt;/p>&lt;p>Exactly! Doing the thing right and doing the right thing.&lt;/p>&lt;p>&lt;em>But doesn&amp;rsquo;t efficiency come from having more capacity?&lt;/em>&lt;/p>&lt;p>No. It&amp;rsquo;s about requiring less effort to achieve an output.&lt;/p>&lt;p>&lt;em>Do you mean working smarter?&lt;/em>&lt;/p>&lt;p>Pretty much. Choosing high-leverage activities and outputs, and ensuring these outputs translate to outcomes.&lt;/p>&lt;p>&lt;em>Sounds great, but what does it mean in practice?&lt;/em>&lt;/p>&lt;p>It means identifying asymmetrical opportunities, and pursuing those.&lt;/p>&lt;p>&lt;em>Not sure I follow&amp;hellip;&lt;/em>&lt;/p>&lt;p>Among the many things we could do, some of them contribute to outcomes considerably more than the effort they require.&lt;/p>&lt;p>&lt;em>Ah! Being smart about what we work on, and ensuring we always work on the most valuable things!&lt;/em>&lt;/p>&lt;p>Yes, but also making sure the way we build things allows us to reduce the effort required to build other things in the future.&lt;/p>&lt;p>&lt;em>You lost me again.&lt;/em>&lt;/p>&lt;p>When we build something, say a software product, depending on how we build it, this will have an impact on its operating effort, and it might or might not allow us to re-use this capability in future products and features that we might build.&lt;/p>&lt;h1 id="companies-re-use-very-little">Companies re-use very little&lt;/h1>&lt;p>&lt;em>This sounds very theoretical, what about an example?&lt;/em>&lt;/p>&lt;p>Of course. Imagine you work for a very large company, like a super bank with more than 300k employees.&lt;/p>&lt;p>&lt;em>Woah, 300k! That&amp;rsquo;d be crazy!&lt;/em>&lt;/p>&lt;p>Indeed! You can easily imagine that 300k employees build many things every year, even after the whole capacity discussion above.&lt;/p>&lt;p>&lt;em>Oh yeah, they can build whatever really!&lt;/em>&lt;/p>&lt;p>And they do. They build a lot of things, including things they already have.&lt;/p>&lt;p>&lt;em>Wait, what? They wouldn&amp;rsquo;t do that, would they?&lt;/em>&lt;/p>&lt;p>They do! All the time! They might have 7 or 8 implementations of the same thing, that all do pretty much the same thing.&lt;/p>&lt;p>&lt;em>Why would they do that?&lt;/em>&lt;/p>&lt;p>Many reasons. Poor visibility, low trust, and dependencies.&lt;/p>&lt;p>&lt;em>So they don&amp;rsquo;t know about these existing implementations, and they don&amp;rsquo;t trust them?&lt;/em>&lt;/p>&lt;p>Yeah, but that&amp;rsquo;s only part of the problem, dependencies are a big deal.&lt;/p>&lt;h2 id="dependencies-prevent-re-using-things">Dependencies prevent re-using things&lt;/h2>&lt;p>&lt;em>What about dependencies? What do you mean?&lt;/em>&lt;/p>&lt;p>Dependencies prevent you from re-using the things you have already built. Imagine you want a banana, and as you pull the banana, it pulls a gorilla, which in turns pulls a jungle.&lt;/p>&lt;p>&lt;em>Are you even serious now?&lt;/em>&lt;/p>&lt;p>It&amp;rsquo;s a silly example, but it works well. You never wanted a gorilla, nor a jungle, but if you want the banana, that&amp;rsquo;s what the banana comes with.&lt;/p>&lt;p>&lt;em>Alright, sure, so what can I do?&lt;/em>&lt;/p>&lt;p>Well, you might give up and deal with the gorilla and the jungle.&lt;/p>&lt;p>&lt;em>That&amp;rsquo;s crazy! No way!&lt;/em>&lt;/p>&lt;p>And that&amp;rsquo;s why usually teams end up re-implementing functionality that already exists. Because of its unwanted dependencies.&lt;/p>&lt;p>&lt;em>Still better than getting the gorilla and the jungle!&lt;/em>&lt;/p>&lt;p>Yeah, but redoing a lot of things many times over and maintaining them is crazy in terms of cost and effort.&lt;/p>&lt;p>&lt;em>I can imagine! Their annual technology budget must be insane! But isn&amp;rsquo;t this a problem all companies have?&lt;/em>&lt;/p>&lt;p>To an extent, yes, it is, although this is a much bigger issue for large companies. Small companies don&amp;rsquo;t usually have many products and software systems, so the amount of re-use is limited anyway. Re-use is still important of course, but this problem scales super-linearly with the size of a company, unfortunately.&lt;/p>&lt;p>&lt;em>Right. So what&amp;rsquo;s the alternative in large companies?&lt;/em>&lt;/p>&lt;p>Modularity.&lt;/p>&lt;h2 id="microservices-dont-enable-re-use">Microservices don&amp;rsquo;t enable re-use&lt;/h2>&lt;p>&lt;em>You mean microservices? I build a microservice, any service can call it, and we achieve re-use.&lt;/em>&lt;/p>&lt;p>No. It&amp;rsquo;s not that simple. It works in theory but not in practice. Microservices are not modifiable or extensible, and you accumulate concentration risk by calling microservices as part of your workflows.&lt;/p>&lt;p>&lt;em>Once again, I think an example would help.&lt;/em>&lt;/p>&lt;p>Oh yeah, sure. Let&amp;rsquo;s say you want to launch a product aimed at increasing retention for businesses. A shop issues reward points, and customers can spend these reward points across different shops. You decide that issuing virtual debit cards that spend these points would be amazing for your users.So, to recap:&lt;/p>&lt;ol>&lt;li>You issue virtual debit cards.&lt;/li>&lt;li>These virtual debit cards can only be accepted in some shops.&lt;/li>&lt;li>These virtual debit cards use reward points that the shops issue.&lt;/li>&lt;li>These virtual debit cards look like cards, but don&amp;rsquo;t go through the VISA or MasterCard schemes, just over HTTP.&lt;/li>&lt;/ol>&lt;p>&lt;em>Okay&amp;hellip;&lt;/em>&lt;/p>&lt;p>And let&amp;rsquo;s imagine that you already have card issuance, card processing, and core banking accounts.&lt;/p>&lt;p>&lt;em>That&amp;rsquo;s right, it doesn&amp;rsquo;t seem like this should take much effort.&lt;/em>&lt;/p>&lt;p>Well, it shouldn&amp;rsquo;t, but it&amp;rsquo;s not that simple.&lt;/p>&lt;p>&lt;em>Why?&lt;/em>&lt;/p>&lt;p>Well, first of all our use case is similar to a normal card transaction, but also different. The way we credit accounts is different. We don&amp;rsquo;t need statements. Statements cannot even be produced in this case, as they&amp;rsquo;re legal documents. We probably don&amp;rsquo;t need to call a 3rd party vendor to check every transaction for sanction lists restrictions in this case. The transactions don&amp;rsquo;t reach us from VISA or MasterCard, but over HTTP.&lt;/p>&lt;p>&lt;em>Oh yeah, I see what you mean. But we&amp;rsquo;d still need to check for fraud attempts, wouldn&amp;rsquo;t we?&lt;/em>&lt;/p>&lt;p>Well, probably, yeah. Can you see that this is a workflow that&amp;rsquo;s very similar to what is there, but we cannot just call a microservice and have our transaction processed?&lt;/p>&lt;p>&lt;em>Totally, it would never work.&lt;/em>&lt;/p>&lt;p>And even if we could, we wouldn&amp;rsquo;t want these marketing-related traffic to be served by the same systems we use for workloads that are way more serious, would be?&lt;/p>&lt;p>&lt;em>Of course, that would be dangerous. This traffic could slow things down or create issues for our real transactions. Imagine how pissed off our customers would be!&lt;/em>&lt;/p>&lt;p>Precisely. And by calling a service outside our boundaries, we need to think about different authentication and authorization mechanisms, how we&amp;rsquo;re going to handle observability, and quite a few other operational aspects.&lt;/p>&lt;p>&lt;em>Yeah, imagine a support rota in such a system!&lt;/em>&lt;/p>&lt;p>Indeed! That&amp;rsquo;d would be hard, figuring out which team to wake up for what, etc.&lt;/p>&lt;p>&lt;em>And it would mean you have dependencies across teams, so if we need something from the card issuance service, we need to ask the team responsible for it to build it.&lt;/em>&lt;/p>&lt;p>Yeah, and they might have other priorities.&lt;/p>&lt;p>&lt;em>Bummer! What can we do instead if microservices don&amp;rsquo;t help here?&lt;/em>&lt;/p>&lt;h1 id="modularity--an-operating-system--self-contained-modules">Modularity = an operating system + self-contained modules&lt;/h1>&lt;p>You need an operating system, and composable modules that are self-contained.&lt;/p>&lt;p>&lt;em>An operating system? Do you mean like Windows?&lt;/em>&lt;/p>&lt;p>Haha, no, it&amp;rsquo;s an analogy. You need something that:&lt;/p>&lt;ol>&lt;li>Allows users to browse which functionality we have already.&lt;/li>&lt;li>Allows users to understand how a piece of functionality works.&lt;/li>&lt;li>Standardizes identity, authentication, authorization, logging, monitoring, and other operational concerns.&lt;/li>&lt;li>Allows users to provision workflows dynamically, on demand, in a self-service fashion.&lt;/li>&lt;li>Lets users configure and combine various pieces of functionality into new products.&lt;/li>&lt;/ol>&lt;p>&lt;em>Wait, isn&amp;rsquo;t this what a Cloud provider does?&lt;/em>&lt;/p>&lt;p>Well, a Cloud provider:&lt;/p>&lt;ol>&lt;li>Allows users to browse which infrastructure resources and software services are available.&lt;/li>&lt;li>Allows users to understand how an infrastructure resource or software service works.&lt;/li>&lt;li>Standardizes identity, authentication, authorization, logging, monitoring, and other operational concerns.&lt;/li>&lt;li>Allows users to provision infrastructure resources and software services dynamically, on demand, in a self-service fashion.&lt;/li>&lt;/ol>&lt;p>&lt;em>Sounds very similar to me.&lt;/em>&lt;/p>&lt;p>Indeed! And the early Cloud providers were born to tackle these issues within the companies that created them! But there is a fundamental difference. Cloud providers provision individual resources and services, leaving you with the effort of configuring how things interact. Say you provision a database and launch a Docker image on ECS, for a microservice that reads from the database&amp;hellip;&lt;/p>&lt;p>&lt;em>Okay&amp;hellip;&lt;/em>&lt;/p>&lt;p>It&amp;rsquo;s your responsibility to allow the microservice to communicate with the database, to create and inject credentials, to open an ingress route that lets HTTP requests reach the microservice.&lt;/p>&lt;p>&lt;em>That&amp;rsquo;s right, that&amp;rsquo;s how Cloud providers work. How is this operating system any different?&lt;/em>&lt;/p>&lt;p>The level of abstraction is different. The modules in our case are more coarse-grained than the ones Cloud providers offer, and we also need to support packaged workflows that use more than 1 module.&lt;/p>&lt;p>&lt;em>So say the gateway, the microservice, and the database are part of the same module?&lt;/em>&lt;/p>&lt;p>Pretty much.&lt;/p>&lt;p>&lt;em>But wouldn&amp;rsquo;t that mean we have dependencies, like the microservice depends on the database?&lt;/em>&lt;/p>&lt;p>Oh, no, here we&amp;rsquo;re talking about behaviour dependencies, not infrastructure dependencies.&lt;/p>&lt;p>&lt;em>What?&lt;/em>&lt;/p>&lt;p>Like, the gorilla is some unwanted business side effect that the banana pulls. It&amp;rsquo;s not an infrastructure dependency. Infrastructure dependencies are good actually, so you don&amp;rsquo;t have to connect the service to the database yourself.&lt;/p>&lt;p>&lt;em>Makes sense. And how does this help with concentration risk and availability problems?&lt;/em>&lt;/p>&lt;p>By allowing users to import a module into their workflows or products.&lt;/p>&lt;p>&lt;em>Like importing and deploying this module ourselves, rather than calling an existing module already deployed somewhere?&lt;/em>&lt;/p>&lt;p>Precisely! This way resources and workloads are isolated, and you avoid both problems.&lt;/p>&lt;p>&lt;em>Nice! And what if I want to change something? If the core accounts module is used as part of the cards processing workflow, we don&amp;rsquo;t want that one, but one that works with rewards points and has no statements. Also, we don&amp;rsquo;t want to screen transactions for issues related to sanctions list, as per our example.&lt;/em>&lt;/p>&lt;p>Yeah, that&amp;rsquo;s where dynamic workflows come in. You can have alternative modules for the same workflow step, and import a different module when you want different behaviour.&lt;/p>&lt;p>&lt;em>Like we&amp;rsquo;d choose to omit the module responsible for statements in this workflow, but include the fraud checker because we want that behaviour, and replace the VISA and MasterCard connectors with an HTTP-based alternative?&lt;/em>&lt;/p>&lt;p>Exactly! Modules should accept configuration options, and be self-contained, meaning that they shouldn&amp;rsquo;t be aware of other modules, but only of events that can happen system-wide.&lt;/p>&lt;h2 id="when-it-comes-to-modules-size-granularity-matters">When it comes to modules, &lt;del>size&lt;/del> granularity matters&lt;/h2>&lt;p>&lt;em>OK, we want to change what happens based on what&amp;rsquo;s needed, on a case-by-case basis. Now that I think about it, could we not just build something that can behave differently based on what&amp;rsquo;s needed? And then pass configuration options to it? Didn&amp;rsquo;t we just say modules should accept configuration options? Based on configuration, we could produce statements or skip that.&lt;/em>&lt;/p>&lt;p>Well, we could, but this approach has many disadvantages over composing self-contained modules.&lt;/p>&lt;p>&lt;em>Such as?&lt;/em>&lt;/p>&lt;p>Well, here are some:&lt;/p>&lt;ul>&lt;li>Higher complexity for the modules, which would need to behave differently based on configuration values. This is a very significant problem. A module that does 2 things is harder to build, test, release, understand, evolve, and maintain, compared to 2 modules that do 1 thing each. And the increase in complexity is also non-linear, so a module that does 4 things is much harder than 4 modules that do 1 thing each.&lt;/li>&lt;li>It introduces dependencies on the “supplier” module. So if you need some different/additional behaviour, you now need to evolve an existing module, rather than composing things. This is not only harder, but typically means asking other teams to do this for you.&lt;/li>&lt;li>It requires modules to be aware of the workflow they&amp;rsquo;re in. Real big issue here. If a module needs to behave differently if another module is in the same workflow, you have workflow-aware modules, so not really modules anymore. Complexity increases significantly, testing becomes incredibly harder.&lt;/li>&lt;li>Modules that do completely different things based on configuration are harder to understand and to use. A user would need to understand what to enable or disable based on the desired workflow.&lt;/li>&lt;/ul>&lt;p>&lt;em>Oh, well, yeah. Like standard Lego blocks, as opposed to configurable and modifiable Lego blocks.&lt;/em>&lt;/p>&lt;p>Yeah, and there&amp;rsquo;s more. The coarser the modules, the less the operating system can infer about them.&lt;/p>&lt;p>&lt;em>What do you mean?&lt;/em>&lt;/p>&lt;p>Let&amp;rsquo;s imagine a product offers an HTTP-based connector, and a Kafka-based connector to ingest data&amp;hellip;&lt;/p>&lt;p>&lt;em>Okay&amp;hellip;&lt;/em>&lt;/p>&lt;p>You could implement this product using multiple modules: 1 for the HTTP-based connector, 1 for the Kafka-based connector, 1 for the processor itself, and some more modules for the outputs e.g. webhooks.&lt;/p>&lt;p>&lt;em>Yeah, makes sense.&lt;/em>&lt;/p>&lt;p>But you could also decide to use a single all-in-one module, or even a couple of coarser modules. Let&amp;rsquo;s say you decide to have a single module that allows to ingest data from HTTP and Kafka, using configuration to decide which one you want: HTTP, Kafka, or both.&lt;/p>&lt;p>&lt;em>Okay&amp;hellip;&lt;/em>&lt;/p>&lt;p>Well, in the first case, the HTTP module and the Kafka module would be declared separately, and the operating system would understand what each of them needs in terms of connectivity, security, permissions, etc. In the second case, if you bundle these capabilities into a single module, the operating system wouldn&amp;rsquo;t be able to understand this, because the distinction of which capability is provided would be determined at runtime, based on the configuration option. So the operating system would need to allow the union set of the connectivity, security, and permissions requirements to this coarser module, even if in a given use case some of these permissions might not be needed at all.&lt;/p>&lt;p>&lt;em>Ouch, that doesn&amp;rsquo;t sound great.&lt;/em>&lt;/p>&lt;p>Well, because it isn&amp;rsquo;t! And if you were to import such a module, you&amp;rsquo;d have to give it all these permissions in any case. Imagine you&amp;rsquo;d like to use this module to allow Kafka data ingestion, you&amp;rsquo;d end up opening HTTP routes even if you don&amp;rsquo;t want HTTP data ingestion.&lt;/p>&lt;p>&lt;em>Yeah, that&amp;rsquo;s definitely a problem.&lt;/em>&lt;/p>&lt;p>It gets even worse. By declaring fine-grained modules, you allow the operating system to reason about changes. If you bundle things into multi-purpose modules, you miss out on this.&lt;/p>&lt;p>&lt;em>Meaning?&lt;/em>&lt;/p>&lt;p>Meaning, as an example, that if a module is declared as stateful, and one as stateless, the operating system could allow to remove the latter without worrying about data retention. If you bundle them into a module that does either, based on a flag, even if you choose the stateless behaviour, this stays within this larger module, so the operating system would now be forced to assume that it could require data retention handling when you unsubscribe from it.&lt;/p>&lt;p>&lt;em>Ah yeah, of course, the operating system wouldn&amp;rsquo;t know in this case!&lt;/em>&lt;/p>&lt;h1 id="simplicity--fewer-moving-parts-more-of-the-same">Simplicity = &lt;del>fewer moving parts&lt;/del> more of the same&lt;/h1>&lt;p>Many people struggle with the concept of simplicity. They tend to say &amp;ldquo;keep it simple&amp;rdquo; every time they don&amp;rsquo;t understand something or instinctively disagree with it. When they see a system made of many parts, they call that complicated.&lt;/p>&lt;p>&lt;em>Isn&amp;rsquo;t that right? Like the fewer the parts the simpler something is?&lt;/em>&lt;/p>&lt;p>Not really. Simplicity means &amp;ldquo;more of the same&amp;rdquo;.&lt;/p>&lt;p>&lt;em>Here we go again, what do you mean?&lt;/em>&lt;/p>&lt;p>Simplicity is not about quantity, but quality. Many simple modules produce simpler systems than the ones you get by using fewer but more complicated modules.&lt;/p>&lt;p>&lt;em>An examples?&lt;/em>&lt;/p>&lt;p>Oh, examples are everywhere. Think about building a 30-stories skyscraper, as opposed to building 30 1-story houses. Which one is simpler?&lt;/p>&lt;p>&lt;em>Mmm&amp;hellip;&lt;/em>&lt;/p>&lt;p>How about a bridge? Is it simpler to produce a 200m-long composite bridge made of 20 10m sections, or a single-piece 200m-long bridge?&lt;/p>&lt;p>&lt;em>I see what you mean.&lt;/em>&lt;/p>&lt;p>Humanity was able to produce large systems a very long time ago, by composing a large number of simple modules. Only very recently we were capable of producing more complicated modules.&lt;/p></description></item><item><title>Demand, capacity, outputs, outcomes, and impact</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-15-demand-capacity-outcomes/</link><pubDate>Sat, 15 Jul 2023 12:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-15-demand-capacity-outcomes/</guid><description>&lt;p>&amp;ldquo;This year we want to double the number of developers!&amp;rdquo;&lt;/p>&lt;p>&amp;ldquo;We can&amp;rsquo;t put UX/UI and QA in each team, it&amp;rsquo;s inefficient!&amp;rdquo;&lt;/p>&lt;p>Sigh, if only reality was so simple. I wonder when screaming your ignorance and incompetence to the world will start getting frowned upon.&lt;/p>&lt;h1 id="the-problem">The problem&lt;/h1>&lt;p>Many managers would tell you that management is an art. But what they mean is that they wing it, without paying any attention to theory or thinking things through.&lt;/p>&lt;p>&amp;ldquo;We need to double the money we make, so we&amp;rsquo;re going to double the number of developers we have!&amp;rdquo;&lt;/p>&lt;p>Now, if you come from another perspective, this sounds insane. How could anyone come up with this? And how could anyone let them make decisions?&lt;/p>&lt;p>In their &amp;ldquo;defense&amp;rdquo;, in many cases the actions they suggest would bring personal benefits, like promotions, more power, or merely attention. So often it&amp;rsquo;s not just a matter of incompetence, but one of integrity. This doesn&amp;rsquo;t come with any judgement, as everybody is self-serving and does what they think is best for them. The problem is that some systems encourage these behaviours, but let&amp;rsquo;s not discuss this today.&lt;/p>&lt;p>So what&amp;rsquo;s the common model so many people in so many companies adopt? Well, they assume that the workforce size translates linearly into the impact e.g. Annual Recurring Revenue (ARR). So if we have 2 salesmen, 4 developers, and are making $1M ARR, if we get to 4 salesmen and 8 developers, we&amp;rsquo;ll be able to make $2M ARR.&lt;/p>&lt;p>Admittedly, this level of delusion is rare, and most people immediately recognize that it makes little sense when they&amp;rsquo;re confronted about it (even if their behaviour often follows this assumption anyway). But other equally simplistic and inadequate models are very popular within the software product development industry. This leads to bad considerations, bad decisions, and typically to bad Outputs, Outcomes, and Impact.&lt;/p>&lt;h1 id="a-better-model-to-think-about-work">A better model to think about work&lt;/h1>&lt;p>Before we can discuss how we can do better, let&amp;rsquo;s remind ourselves of some concepts.&lt;/p>&lt;h2 id="some-definitions">Some definitions&lt;/h2>&lt;ul>&lt;li>Work: work is anything a person, a team, or a Company does, according to what the company expects of them. Meetings, writing code, answering the phone, reports, etc. Checking your personal Facebook page is not work.&lt;/li>&lt;li>Employee: a person working for the Company. For this analysis we limit this to Line Employees, the ones that generate Capacity, as opposed to Managers.&lt;/li>&lt;li>Workforce Size: the number of Employees contributing Capacity.&lt;/li>&lt;li>Demand (D): all the things a Company asks its employees to work on.&lt;/li>&lt;li>Outputs: the stuff you build or produce as a Company, a team, or an employee.&lt;/li>&lt;li>Capacity (C): how much a unit of work allocation contributes to the Outputs. Could be used for an employee, but makes more sense for a team or a company. It&amp;rsquo;s the Outputs progressed over a time window.&lt;/li>&lt;li>Outcomes: the amount of value a Company delivers to its market. It&amp;rsquo;s the value customers and users get from the company&amp;rsquo;s products and services. Note that Outputs can improve Outcomes, but they can as easily harm them.&lt;/li>&lt;li>Impact: the amount of value a Company captures from its market. It&amp;rsquo;s the amount of resources (money, attention, loyalty, etc.) the company can get from its market. Outputs cause changes in Outcomes, which cause changes in Impact. Note that Outcomes might improve or harm Impact.&lt;/li>&lt;li>Wealth: the difference between the value delivered minus the resources absorbed. It&amp;rsquo;s Outcomes minus Impact.&lt;/li>&lt;li>Company: a social system whose purpose is to produce and redistribute Wealth in its target market, indefinitely over time.&lt;/li>&lt;li>Busy: the state of an Employee that&amp;rsquo;s carrying over Work.&lt;/li>&lt;li>Idle: the state of an Employee that is not carrying over Work.&lt;/li>&lt;li>Utilization (U): the ratio between busy time and idle time. Usually applied to Employees, but could be considered for teams as well. So an employee that&amp;rsquo;s constantly busy will have a Utilization of 100%, while a constantly idle Employee will have a Utilization of 0%.&lt;/li>&lt;li>Productivity: the percentage of Work which contributes to the Capacity. If an Employee works hard, but most of this work does not contribute to the Outputs, then that Employee is not very productive. Same for a team or a company.&lt;/li>&lt;li>Value-Added Work: any work that:&lt;ul>&lt;li>A customer is willing to pay for.&lt;/li>&lt;li>Transforms the product.&lt;/li>&lt;li>Is done right the first time.&lt;/li>&lt;/ul>&lt;/li>&lt;li>Non-Value-Added Work (or Waste): work that consumes resources, but does not add value to the product or service. Mind that some Non-Value-Added Work (or waste) might be necessary e.g. contracts, regulatory requirements, etc.&lt;/li>&lt;li>Workflow: the sequence of industrial, administrative, or other processing steps through which a piece of work passes from initiation to completion.&lt;/li>&lt;li>Processing: the actions performed to progress a piece of work.&lt;/li>&lt;li>(Processing) Step: a phase of the Workflow for a piece of work. Actions that are intended to design, transform, inspect, or deliver a product.&lt;/li>&lt;li>Throughput (TP): the quantity of Outputs delivered in the unit of time, averaged out. Assuming all work items are the same, it&amp;rsquo;s the number of finished work items you delivered in the time window you&amp;rsquo;re considering.&lt;/li>&lt;li>Queue (Q): a backlog of pieces of work a Processing Step must process before it can process other work.&lt;/li>&lt;li>Queue Time (QT): the amount of time a piece of work spends waiting before getting processed.&lt;/li>&lt;li>Processing Time (PT): the amount of time a piece of work spends getting processed.&lt;/li>&lt;li>Total Processing Time (TPT): the total amount of time a piece of work spends getting processed along its whole workflow.&lt;/li>&lt;li>Total Queue Time (TQT): the total amount of time a piece of work spends waiting along its whole workflow.&lt;/li>&lt;li>Cycle Time (CT): it&amp;rsquo;s the amount of time it takes for the Workflow for a piece of work to complete. From when we start working on it, to when it&amp;rsquo;s finished and delivered. It&amp;rsquo;s the Total Processing Time + the Total Queue Time.&lt;/li>&lt;li>Lead Time (LT): it&amp;rsquo;s the amount of time it takes from a piece of work to go from desired to finished and delivered. It&amp;rsquo;s that piece of work&amp;rsquo;s Cycle Time, plus the Queue Time before first Step of the Workflow can commence.&lt;/li>&lt;li>Work In Progress (WIP): the amount of work in progress at a given time. So if at any given time you have 4 pieces of work that started their Workflow but haven&amp;rsquo;t finished it, your Work In Progress is 4 at that time.&lt;/li>&lt;li>&lt;a href="https://flowchainsensei.wordpress.com/2018/06/22/the-folks-that-matter/">Folks That Matter&lt;/a>: every person whose needs we need to meet, for otherwise this will have negative consequences for the company. Folks That Matter can be external e.g. customers, users, suppliers, regulators, shareholders, investors, etc. or internal e.g. finance, legal, engineers, product people, sales folks, etc. Kudos to &lt;a href="https://flowchainsensei.wordpress.com/">Bob Marshall&lt;/a> for coming up with this awesome concept.&lt;/li>&lt;li>Quality: the degree to which a product meets the needs of all the Folks That Matter.&lt;/li>&lt;li>External Quality: like Quality, but only when considering the needs of the external Folks That Matter.&lt;/li>&lt;li>Internal Quality: like Quality, but only when considering the needs of the internal Folks That Matter.&lt;/li>&lt;li>Failure Demand: it&amp;rsquo;s all the work we must do because we failed to fully meet the needs of some of the Folks That Matter.&lt;/li>&lt;li>Value Demand: it&amp;rsquo;s the work we must do because it delivers value to somebody. It&amp;rsquo;s the Demand, minus the Failure Demand.&lt;/li>&lt;li>Cost of Delay: the cost of delaying some piece of work. This is typically expressed in &amp;ldquo;bucks a day&amp;rdquo;, and in terms of how it changes over time.&lt;/li>&lt;li>Efficiency: how much Capacity we can generate, for a given number of Employees. All else being equal, if it takes 5 people 1 week to produce the same Output 10 people would produce in a week by using another approach, then the first approach is twice as efficient. Efficiency concerns itself with doing the thing right.&lt;/li>&lt;li>Effectiveness: it concerns itself with doing the right thing. Given a fixed capacity, how much we can make a positive difference in Outcomes and Impact, by producing Outputs. So whenever we use our hard-earned Capacity to produce Outputs, but then those Outputs do not produce desirable changes in Outcomes and Impact, we&amp;rsquo;re not being effective. Failure Demand subtracts from Effectiveness, as it requires us to use our Capacity to carry over work which wouldn&amp;rsquo;t be needed in the first place, if we hadn&amp;rsquo;t failed at meeting the needs of all the Folks That Matter.&lt;/li>&lt;/ul>&lt;p>Now that we can name things, let&amp;rsquo;s attempt a better way of thinking about work. This model is not &amp;ldquo;true&amp;rdquo;, because the point of a model is not to correct, but to be useful. Therefore, even this model omits large amounts of details and information that are not deemed useful in order to facilitate making predictions and decisions within the context of software product delivery.&lt;/p>&lt;h2 id="the-context">The context&lt;/h2>&lt;p>A company:&lt;/p>&lt;ul>&lt;li>Has a number of Employees.&lt;/li>&lt;li>Has Employees working in a certain way.&lt;/li>&lt;li>Has a level of Capacity, deriving from the quality and quantity of its Employees, and the way the work.&lt;/li>&lt;li>Has Demand as what Outputs to deliver to address the needs of the Folks That Matter.&lt;/li>&lt;li>Invests its Capacity over time to produce Outputs and address Demand.&lt;/li>&lt;li>Sees changes in Outcomes and Impact from the Outputs it delivers. These can be positive, negative, or completely unaffected.&lt;/li>&lt;/ul>&lt;h2 id="desirable-properties">Desirable properties&lt;/h2>&lt;p>Many companies implicitly assume that their Capacity is proportional to their number of Employees. This applies overall, or within the context of a given project.&lt;/p>&lt;p>In terms of desirable properties, in order of importance, we got:&lt;/p>&lt;ol>&lt;li>High Quality: our efforts meet the needs of all the Folks That Matter.&lt;/li>&lt;li>High Effectiveness: we only work on what moves the needle. Our Outputs produce desirable Outcomes and Impact.&lt;/li>&lt;li>Low Lead Time: once we set our mind and want something, we can deliver it quickly. This is key to minimize Cost of Delay, and to stay reactive to risks and opportunities.&lt;/li>&lt;li>Even flow: we deliver things often in increments, as opposed to staggering our value delivery so that things happen in bulk. This way we deliver value early, and we get earlier feedback on our Quality.&lt;/li>&lt;li>High Throughput (or Capacity): within a time window, we deliver a lot of Outputs.&lt;/li>&lt;li>High Efficiency: we can generate more Capacity with fewer Employees.&lt;/li>&lt;/ol>&lt;p>It&amp;rsquo;s important to keep in mind that these things need to be optimised in order, and for the long term. If by doing something we become more efficient but less effective, or our efficiency collapses after a quarter, that&amp;rsquo;s not something we should do.&lt;/p>&lt;h2 id="how-to-maximise-outputs">How to maximise outputs&lt;/h2>&lt;p>There&amp;rsquo;s a lot to talk about in terms of how to design a system of work that caters for the desired properties above, in a way that&amp;rsquo;s resilient and sustainable.&lt;/p>&lt;p>But alas this is a blog post, not a book!&lt;/p>&lt;p>So instead of attempting to discuss a good model now (I&amp;rsquo;ll try in another post), I&amp;rsquo;ll leave it to you to consider how you work and what assumptions are behind this way of working:&lt;/p>&lt;ol>&lt;li>Does your software product development workflow involve more than 1 step?&lt;ul>&lt;li>If so, how many? Why? Which ones?&lt;/li>&lt;li>Do you have Specification -&amp;gt; Design -&amp;gt; Development -&amp;gt; QA -&amp;gt; Release?&lt;/li>&lt;li>Are these steps performed by the same people or by different people?&lt;/li>&lt;li>What happens when Development is &amp;ldquo;finished&amp;rdquo; for a piece of work, but QA is not free to start working on it? To work in progress? To lead time?&lt;/li>&lt;li>What happens when Development is &amp;ldquo;finished&amp;rdquo; for a piece of work, but nothing is in the queue because &amp;ldquo;Design&amp;rdquo; is lagging behind? To work in progress? To lead time?&lt;/li>&lt;li>What are the assumptions behind this way of structuring a workflow?&lt;/li>&lt;/ul>&lt;/li>&lt;li>For a single step e.g. Development or QA:&lt;ul>&lt;li>Do you think that adding more people will increase the capacity of that step?&lt;/li>&lt;li>What are the assumptions behind this if you do? Which ones if you don&amp;rsquo;t?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Do your teams specialise in or own an area?&lt;ul>&lt;li>Why?&lt;/li>&lt;li>What are the consequences of specialisation?&lt;/li>&lt;li>What happens when you split a piece of work across teams?&lt;/li>&lt;li>How do you structure your on-call for customer support?&lt;/li>&lt;li>What assumptions are behind this way of working?&lt;/li>&lt;/ul>&lt;/li>&lt;li>How much of your Demand is Failure Demand?&lt;ul>&lt;li>Everything that wouldn&amp;rsquo;t happen if your products and services fully met the needs of all the Folks That Matter.&lt;/li>&lt;li>How often are you reactive vs strategic in terms of what your teams work on?&lt;/li>&lt;/ul>&lt;/li>&lt;li>How often do you release?&lt;ul>&lt;li>What happens if we need something really urgently?&lt;/li>&lt;li>Do we stop people and ask them to switch context? What happens to the work they were doing before? What happens when you switch context?&lt;/li>&lt;li>How likely is that a release will impact your users negatively (bugs, slowness, unavailability, etc.)? Why?&lt;/li>&lt;li>What assumptions are behind your release cadence?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Do your people work individually or together as a team?&lt;ul>&lt;li>Do you have individual tasks? Who decides who works on what? Based on what? What if this decision is not optimal?&lt;/li>&lt;li>Do you have individual rewards e.g. bonuses? What&amp;rsquo;s the implicit message you&amp;rsquo;re sending your employees?&lt;/li>&lt;li>What does teamwork mean to you? Do your &amp;ldquo;teams&amp;rdquo; behave like street cleaners, or like basketball/football/military teams? What&amp;rsquo;s the difference?&lt;/li>&lt;li>What assumptions are behind the way your teams work?&lt;/li>&lt;/ul>&lt;/li>&lt;li>What does your company believe about speed and quality? What about quality and cost?&lt;ul>&lt;li>Do people see these as a trade-off?&lt;/li>&lt;li>What&amp;rsquo;s the effect quality has on the time and cost of performing an activity?&lt;/li>&lt;li>What effect does it have on Failure Demand?&lt;/li>&lt;li>What does this mean for cost?&lt;/li>&lt;/ul>&lt;/li>&lt;li>What&amp;rsquo;s the relationship between the Utilization of the members of a team, and the effectiveness of that team?&lt;ul>&lt;li>What effect does high Utilization have on throughput? On lead time? On cycle time? On quality?&lt;/li>&lt;li>What happens when a road gets to 100% Utilization? What happens to lead time and throughput?&lt;/li>&lt;/ul>&lt;/li>&lt;li>What percentage of the things your company decides to build are valuable to your customers?&lt;ul>&lt;li>How do you know?&lt;/li>&lt;li>How can you not know?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Which incentives does your company use on its employees?&lt;ul>&lt;li>Why?&lt;/li>&lt;li>Which behaviours do they encourage?&lt;/li>&lt;li>Are these behaviours desirable?&lt;/li>&lt;/ul>&lt;/li>&lt;li>What other implications do your ways of working have on the employees?&lt;ul>&lt;li>How do you think people feel, given the ways of working and the incentives?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Do you use GitFlow?&lt;ul>&lt;li>Why?&lt;/li>&lt;li>Do you see any queues, handovers, and waste?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Do managers in your company talk about these things?&lt;ul>&lt;li>Why? Why not?&lt;/li>&lt;li>What are the implicit assumptions?&lt;/li>&lt;li>Do they manage the work or how the work works? Why? What are the consequences?&lt;/li>&lt;/ul>&lt;/li>&lt;li>Why do companies want to double their number of developers?&lt;ul>&lt;li>What are the implicit assumptions here?&lt;/li>&lt;li>Why might this not be a great idea?&lt;/li>&lt;li>What should they be doing instead?&lt;/li>&lt;/ul>&lt;/li>&lt;/ol></description></item><item><title>Management by Objectives is wishful thinking</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-09-management-by-objectives-is-wishful-thinking/</link><pubDate>Sun, 09 Jul 2023 19:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-09-management-by-objectives-is-wishful-thinking/</guid><description>&lt;p>&amp;ldquo;This quarter we want revenue up by 10%&amp;rdquo; - but is it really that simple? Or is Management by Objectives a travesty?&lt;/p>&lt;p>There are 2 ways of managing.&lt;/p>&lt;p>One is hard, requires knowledge and understanding, and will expose your mistakes when you make any.&lt;/p>&lt;p>The other one doesn&amp;rsquo;t work, is delusional, and betrays both the top and the bottom. But it&amp;rsquo;s easy, comes at zero personal risk and effort, and works the same across any company, without any prerequisite knowledge.&lt;/p>&lt;p>Guess which one is so ubiquitous in the software product development industry that many people assume it&amp;rsquo;s the only possibility? Which one do you think is pushed by the certification industry and the business schools? Which one do you reckon is hinted at or directly asked for in almost every management job spec?&lt;/p>&lt;p>Thought so. As sad as it still feels after so many years, let&amp;rsquo;s look at what this nonsensical idea about management looks like, why it&amp;rsquo;s appealing even if delusional, and what the sane alternative works like.&lt;/p>&lt;h2 id="management-by-objectives">Management by Objectives&lt;/h2>&lt;p>The main idea behind Management by Objectives is that, as a manager, you manage people by assigning objectives to them. Their goal becomes then meeting those objectives. You, the manager, are part of &amp;ldquo;the business&amp;rdquo;, while &amp;ldquo;they&amp;rdquo;, the makers, are part of the workforce, not to be trusted or consulted, but to be directed and controlled. Because without control, people will skive off.&lt;/p>&lt;p>In virtue of the extensive knowledge and understanding you were imparted in business school, you can join any company, and come up with objectives for the people and the processes that you manage.&lt;/p>&lt;p>M: &amp;ldquo;How many screws do we produce per day?&amp;rdquo;&lt;/p>&lt;p>E: &amp;ldquo;400,000&amp;rdquo;&lt;/p>&lt;p>M: &amp;ldquo;So by the end of Q2 you&amp;rsquo;ll need to be producing 440,000 screws a day, a 10% increase&amp;rdquo;&lt;/p>&lt;p>Some employees might push back to this nonsense, but most are already too jaded to even try, and even the ones that do soon give up or get forced out. So 440,000 becomes the objective.&lt;/p>&lt;p>How though? Why? And why now?&lt;/p>&lt;p>Because if producing 440,000 screws a day was simply a matter of willing to do so, why were we not doing it this quarter? If it was that easy, why would we not do 500,000 instead?&lt;/p>&lt;p>What follows here is usually 1 of these 3 scenarios:&lt;/p>&lt;ol>&lt;li>The employees fail to achieve the objective, and get blamed for it.&lt;/li>&lt;li>The employees achieve the objective, at the cost of screwing up important things that were not explicitly included in the targets.&lt;/li>&lt;li>The employees achieve the target, by improving the way the work works.&lt;/li>&lt;/ol>&lt;p>Can you guess what&amp;rsquo;s common about all 3 possible outcomes? The manager risks nothing, and if things turn out okay they&amp;rsquo;ll be hailed as the hero of the situation.&lt;/p>&lt;p>So what happens in these scenarios? Let&amp;rsquo;s have a look.&lt;/p>&lt;h3 id="scenarios">Scenarios&lt;/h3>&lt;h4 id="failing-to-achieve-the-objective">Failing to achieve the objective&lt;/h4>&lt;p>When the employees fail to achieve the objective, nobody questions the objective itself. Nobody goes like &amp;ldquo;uh, how do we know this was even possible?&amp;rdquo;, the blame inevitably falls on the makers. &amp;ldquo;The business asked you to achieve X, and you failed, so you&amp;rsquo;re bad. Get your game together, or there will be repercussions&amp;rdquo;.&lt;/p>&lt;p>What effect does this have on the makers? Well, the best leave pretty soon. They get what game they&amp;rsquo;re forced to play, have options, and decide to take their expertise and domain knowledge somewhere else. They are typically addressed by &amp;ldquo;the business&amp;rdquo; as &amp;ldquo;B players&amp;rdquo;, despite their history of amazing performance reviews.&lt;/p>&lt;p>The rest? Well, the rest adjust. If meeting the objective is what matters, they&amp;rsquo;ll ensure the objective is met. At least on paper. Story points? Lines of code? Shipping? Test coverage? Any metric can be gamed, or achieved at the expense of something important, but not measured or measurable.&lt;/p>&lt;h4 id="achieving-the-objective-at-the-cost-of-harming-other-important-things">Achieving the objective at the cost of harming other important things&lt;/h4>&lt;p>Sometimes employees achieve the objective by harming other important things that were not kept under observation. The manager is hailed as hero, and the makers as well. Sounds great? Well, not really. What&amp;rsquo;s likely to happen the next time? The manager is still euphoric about the &amp;ldquo;personal achievements&amp;rdquo; recently conquested, and the employees have found a path to success theatre.&lt;/p>&lt;p>M: &amp;ldquo;So this time we&amp;rsquo;re aiming at 484,000, another 10% increase&amp;rdquo;.&lt;/p>&lt;p>E: &amp;ldquo;Uh, uh, yes sir (here we go again)&amp;rdquo;.&lt;/p>&lt;h4 id="the-employees-achieve-the-objective-due-to-improving-how-the-work-works">The employees achieve the objective, due to improving how the work works&lt;/h4>&lt;p>First of all, let me say that this is exceedingly infrequent. Employees rarely have the autonomy to try anything new, not to mention anything substantial, which involves processes and spans across departments.&lt;/p>&lt;p>But hey, sometimes miracles happen. And when they do, guess who gets all the merit? The manager, of course.&lt;/p>&lt;p>M: &amp;ldquo;See? I pushed them hard and improved things! Wait until you see where we&amp;rsquo;re going to be the next quarter!&amp;rdquo;&lt;/p>&lt;h3 id="so-what">So what?&lt;/h3>&lt;p>Well, you have managers who know and risk nothing, and get all the recognition. From executives to line managers, through middle-managers, the song remains the same. No accountability whatsoever.&lt;/p>&lt;p>In the meantime, employees get blamed, become disenchanted, burn out, or leave altogether. Quality and reputation inevitably go down, the best talent leaves the company, and the message for the makers is &amp;ldquo;figure out the quickest way to get into management, so that you can do nothing and risk nothing, while being paid more and celebrated for success that wasn&amp;rsquo;t your own&amp;rdquo;.&lt;/p>&lt;h2 id="management-by-means">Management by Means&lt;/h2>&lt;p>On the opposite scale of convenience lies Management by Means - like Deming used to call it. Bad news though, to figure out means of improving capacity you need to deeply understand many things, including variation, systems, and people. More importantly, though, you need to be intimate with the work and how the work works.&lt;/p>&lt;p>Compare the following statements:&lt;/p>&lt;p>M1: &amp;ldquo;The objective is +20% story points completed every 2 weeks&amp;rdquo;.&lt;/p>&lt;p>M2: &amp;ldquo;By redesigning the teams to be equivalent and cross-functional, and to pull work from a prioritized queue of initiatives, we hope that over the next quarter we&amp;rsquo;ll improve our effectiveness and resilience&amp;rdquo;.&lt;/p>&lt;p>Which one do you think has an easier job? Whose mistakes will be more evident, when any? Which manager needs to know more about software development, people, variation, processes, and software development ways of working? Which manager could more easily do their work unaltered in a completely different industry?&lt;/p>&lt;p>And yet, this is exactly what management is. Not managing people, but managing how the work works. It&amp;rsquo;s about providing insights, and directing the continuous improvement efforts, not obsessing over measurements and targets, but focusing on assessing the current demand and capacity, and continuously improving the capacity to better meet the demand.&lt;/p>&lt;h1 id="so-what-1">So what?&lt;/h1>&lt;p>Well, so be careful about who you hire, and recognize the people around you for who they really are. When you work for a company, don&amp;rsquo;t give in to the incessant pressure to conform with the countless instruments of management by objectives.&lt;/p>&lt;p>Because regardless of names, whether it&amp;rsquo;s OKRs, KPIs, individual goals, team goals, or performance metrics, these are tools in the hands of people who know nothing and are good for nothing. Setting a goal and &amp;ldquo;letting people figure it out&amp;rdquo; is not management, it&amp;rsquo;s wishful thinking and delusion. Worse, it&amp;rsquo;s exploitation of the fools at the top of the company.&lt;/p>&lt;p>Management is about identifying purpose, assessing demand and capacity, and then continuously improving capacity by leveraging knowledge and understanding about how the work works.&lt;/p></description></item><item><title>Traits of great employees</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-02-traits-of-great-employees/</link><pubDate>Sun, 02 Jul 2023 14:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-02-traits-of-great-employees/</guid><description>&lt;p>Companies always talk about wanting to hire A players and great talent. But what does great talent mean? How can you tell whether somebody will be great at your company?&lt;/p>&lt;p>Companies employ people. In collaborative knowledge work, like software product development, employees are the most important asset for any company. So it&amp;rsquo;s only natural that companies try hard to secure the best talent. And yet many companies don&amp;rsquo;t ask themselves what great talent means.&lt;/p>&lt;p>Also, the typical underlying assumption is that people own their performance, that the extent somebody contributes to a company depends mostly or entirely on the employee. But is this really true? Can anyone claim that their impact and contribution does not depend on their containing system of work, like the company, the team, and the processes?&lt;/p>&lt;p>Everybody does what they think is best for themselves, no exceptions. So if the same person were to be part of 2 different companies which incentivize different behaviours, it&amp;rsquo;s pretty much guaranteed that that person would behave differently, according to what&amp;rsquo;s best for them in each context.&lt;/p>&lt;h3 id="so-what">So what?&lt;/h3>&lt;p>Well, what I&amp;rsquo;m trying to say is that people don&amp;rsquo;t have intrinsic qualities that stay constant regardless of your company culture, incentives, and processes.&lt;/p>&lt;p>Which is both good and bad news. If you structure the right incentives, design effective processes, and end up producing a great culture, then every employee you hire will shine. But conversely, if you have ineffective incentives, culture, and processes, even A players and great talent will adjust accordingly.&lt;/p>&lt;p>Should we just give up interviewing for personality then? Well, not really. While it&amp;rsquo;s true that the incentives and the system will deeply affect an employee&amp;rsquo;s behaviour and performance, people enjoy different things, so the key is to find employees that are happy and passionate about working in a company that values the behaviours you want to see.&lt;/p>&lt;h3 id="what-great-talent-looks-like">What great talent looks like&lt;/h3>&lt;p>What follows is a collection of traits, in no particular order, that a great employee should embody.&lt;/p>&lt;p>Your goal as a company is to structure incentives, ways of working, and processes to bring out these qualities in the people you staff. Remember that if you incentivize the opposite, nobody will exhibit these behaviours.When you interview candidates, you&amp;rsquo;re looking for people that are passionate and excited about working in an environment that promotes these traits.&lt;/p>&lt;h4 id="ownership">Ownership&lt;/h4>&lt;p>Ownership is the willingness to give it all and contribute to something all around. Taking responsibility for work, dealing with uncertainty, prototyping, asking for feedback, learning new things, and doing everything needed to move things forward.&lt;/p>&lt;p>It is &lt;em>NOT&lt;/em> about working hard, commitments, meeting arbitrary deadlines, or caring as much as the founders about their business.&lt;/p>&lt;h5 id="what-happens-when-its-missing">What happens when it&amp;rsquo;s missing&lt;/h5>&lt;ul>&lt;li>You cannot scale your business, because you constantly need to break down work, chase people, review progress, and check quality.&lt;/li>&lt;li>The more you hire the busier you get.&lt;/li>&lt;li>Problems arise from areas nobody takes care of.&lt;/li>&lt;/ul>&lt;h5 id="what-incentives-affect-it">What incentives affect it&lt;/h5>&lt;ul>&lt;li>How much freedom and responsibility you give to your employees. If you treat people like kids they&amp;rsquo;ll behave accordingly.&lt;/li>&lt;li>Anything that promotes a transactional culture destroys ownership. Individual bonuses are rewards are a very dangerous incentive in this sense.&lt;/li>&lt;li>Defining KPIs and performance dimensions, implicitly also stating that anything else doesn&amp;rsquo;t matter.&lt;/li>&lt;/ul>&lt;h4 id="outcome-orientation">Outcome orientation&lt;/h4>&lt;p>Being outcome oriented means starting from the desired outcomes, and walking back to the technologies, the design, the approaches, and the roadmap.&lt;/p>&lt;p>It is &lt;em>NOT&lt;/em> about not caring about technical excellence or disregarding quality.&lt;/p>&lt;h5 id="what-happens-when-its-missing-1">What happens when it&amp;rsquo;s missing&lt;/h5>&lt;ul>&lt;li>Employees choose what they&amp;rsquo;re familiar and comfortable with, which often might lead to undesired outcomes.&lt;/li>&lt;li>There&amp;rsquo;s a lot of effort and being busy, but very little progress, with 2 steps forward and 3 steps back.&lt;/li>&lt;/ul>&lt;h5 id="what-incentives-endangers-it">What incentives endangers it&lt;/h5>&lt;ul>&lt;li>Social or career dangers when mistakes are made is a sure way to discourage people from thinking and trying things they&amp;rsquo;re not comfortable with.&lt;/li>&lt;li>Rewarding outputs e.g. story points, features, shipping, meeting arbitrary deadlines, finishing the tasks in a Sprint, etc.&lt;/li>&lt;li>Rigid roles and responsibilities.&lt;/li>&lt;li>Processes involving handovers, reviews, and multiple teams.&lt;/li>&lt;/ul>&lt;h4 id="long-term-thinking">Long-term thinking&lt;/h4>&lt;p>Long-term thinking is about considering the effects of decisions not only in terms of the near future, but along a longer time horizon. It&amp;rsquo;s thinking about the interdependence of people, technologies, processes, and divisions. It&amp;rsquo;s considering maintenance and evolution costs.&lt;/p>&lt;p>It is &lt;em>NOT&lt;/em> about lack of action, ivory tower discussions, not shipping anything ever.&lt;/p>&lt;h5 id="what-happens-when-its-missing-2">What happens when it&amp;rsquo;s missing&lt;/h5>&lt;ul>&lt;li>You might meet any arbitrary checkpoint on the way, and fail to reach your destination.&lt;/li>&lt;li>Scaling might prove fatal to your business.&lt;/li>&lt;li>Firefighting becomes prevalent, leaving very little capacity for anything else.&lt;/li>&lt;/ul>&lt;h5 id="what-incentives-endangers-it-1">What incentives endangers it&lt;/h5>&lt;ul>&lt;li>Artificial deadlines.&lt;/li>&lt;li>Pushed-based work, where management assigns work to people or teams.&lt;/li>&lt;li>A culture of keeping busy and &amp;ldquo;get shit done&amp;rdquo;, as opposed to &amp;ldquo;think things through&amp;rdquo;.&lt;/li>&lt;li>Rewarding outputs and meeting quarterly goals.&lt;/li>&lt;li>Not distinguishing speed of action from speed of progress.&lt;/li>&lt;/ul>&lt;h4 id="team-spirit">Team spirit&lt;/h4>&lt;p>Team spirit is about being part of a collective effort, without the need to lead or get personal recognition in every situation.&lt;/p>&lt;p>It&amp;rsquo;s &lt;em>NOT&lt;/em> about sacrificing your own goals and aspirations, being okay when your strengths are not recognized, or not wanting career advancement.&lt;/p>&lt;h5 id="what-happens-when-its-missing-3">What happens when it&amp;rsquo;s missing&lt;/h5>&lt;ul>&lt;li>Your employees compete with each other for rewards, attention, and recognition. They can sabotage others&amp;rsquo; efforts to come on top.&lt;/li>&lt;li>People don&amp;rsquo;t teach each other what they know.&lt;/li>&lt;li>Folks don&amp;rsquo;t help each other, so if somebody is struggling or behind with something, that something typically ends up failing.&lt;/li>&lt;/ul>&lt;h5 id="what-incentives-endangers-it-2">What incentives endangers it&lt;/h5>&lt;ul>&lt;li>Individual tasks and performance reviews.&lt;/li>&lt;li>Individual bonuses and rewards.&lt;/li>&lt;li>Failing to recognize contributions to the way a team operates and efforts in helping others conveys the message that the only thing that matter is direct personal outputs.&lt;/li>&lt;/ul>&lt;h4 id="willingness-to-speak-up-and-challenge-ideas">Willingness to speak up and challenge ideas&lt;/h4>&lt;p>People who are willing to speak up and challenge ideas take a personal risk - social or career - and warn others about problems they see or disagreements they have.&lt;/p>&lt;p>This is &lt;em>NOT&lt;/em> being disrespectful of others, expecting for others to always listen, wanting to always get it your way.&lt;/p>&lt;h5 id="what-happens-when-its-missing-4">What happens when it&amp;rsquo;s missing&lt;/h5>&lt;ul>&lt;li>Mediocrity reigns supreme, as many good ideas stay unheard and unconsidered.&lt;/li>&lt;li>Meetings see 1-2 people speaking and the vast majority of the rest being silent, becoming useless.&lt;/li>&lt;li>Employees can feel useless and leave.&lt;/li>&lt;/ul>&lt;h5 id="what-incentives-endangers-it-3">What incentives endangers it&lt;/h5>&lt;ul>&lt;li>Making fun or punishing people when they propose something silly.&lt;/li>&lt;li>Separating the workforce in a group that&amp;rsquo;s supposed to think and another that&amp;rsquo;s supposed to do as they&amp;rsquo;re told.&lt;/li>&lt;li>A culture of efficiency and keeping busy.&lt;/li>&lt;/ul></description></item><item><title>A categorization of what's knowable</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-01-a-categorization-of-whats-knowable/</link><pubDate>Sat, 01 Jul 2023 10:30:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-07-01-a-categorization-of-whats-knowable/</guid><description>&lt;p>We often use different words for what we know, like in data-engineering, information systems, and knowledge workers. But do we really understand their meaning and implications?&lt;/p>&lt;h3 id="so-what-are-these-categories">So what are these categories?&lt;/h3>&lt;p>&lt;a href="https://en.wikipedia.org/wiki/Russell_L._Ackoff">Russell Ackoff&lt;/a> used to say and teach that the content of the human mind can be classified into five categories:&lt;/p>&lt;ol>&lt;li>Data: individual observed phenomena&lt;/li>&lt;li>Information: data that&amp;rsquo;s been processed and organized to be useful; provides answers to &amp;ldquo;who&amp;rdquo;, &amp;ldquo;what&amp;rdquo;, &amp;ldquo;where&amp;rdquo;, and &amp;ldquo;when&amp;rdquo; questions&lt;/li>&lt;li>Knowledge: applied information and data, answers &amp;ldquo;how&amp;rdquo; questions&lt;/li>&lt;li>Understanding: appreciation of &amp;ldquo;why&amp;rdquo;&lt;/li>&lt;li>Wisdom: evaluated understanding&lt;/li>&lt;/ol>&lt;h4 id="data">Data&lt;/h4>&lt;p>A data point is a single observed phenomenon. Data has no significance beyond its existence. It doesn&amp;rsquo;t have meaning of itself.&lt;/p>&lt;p>Data is everywhere, with phenomena continuously happening, being observed, and being recorded.&lt;/p>&lt;h5 id="examples">Examples&lt;/h5>&lt;ul>&lt;li>It&amp;rsquo;s raining.&lt;/li>&lt;li>The sky is blue.&lt;/li>&lt;li>There&amp;rsquo;s a drop on blood on the floor.&lt;/li>&lt;/ul>&lt;h4 id="information">Information&lt;/h4>&lt;p>Information is data that&amp;rsquo;s been processed and organized to be useful. Information gives meaning to data, by creating relations between data points, which typically involves tagging, categorizing, and aggregating. Mind that the meaning information gives to data isn&amp;rsquo;t intrinsically useful.&lt;/p>&lt;p>Information needs to be actively produced, by processing data based on a &amp;ldquo;who&amp;rdquo;, &amp;ldquo;what&amp;rdquo;, &amp;ldquo;where&amp;rdquo;, or &amp;ldquo;when&amp;rdquo; question, and can be inferred from data e.g. when we produce an average.&lt;/p>&lt;h5 id="examples-1">Examples&lt;/h5>&lt;ul>&lt;li>The average height of Italian men is 175 cm.&lt;/li>&lt;li>Yesterday Bob woke up at 6 a.m., went to the gym at 6.30 a.m., came back home at 8 a.m., went to work at 9 a.m., and came back home at 7 p.m.&lt;/li>&lt;li>The rainfall in Paris was 641 mm this year.&lt;/li>&lt;/ul>&lt;h4 id="knowledge">Knowledge&lt;/h4>&lt;p>Knowledge is the appropriate collection of information, such that it&amp;rsquo;s intent is to be useful. This hierarchy adopts knowledge as know-how, or skill, rather than knowledge in the sense of the know-that of propositional knowledge.&lt;/p>&lt;p>Knowledge as know-how transforms information into instructions. It makes control of a system possible, in virtue of knowing how the system works. To control a system is to make it work efficiently. Knowledge is gained by analysis, by taking something apart, figuring out how its parts work, and by then putting together the knowledge of the parts into the knowledge of the whole.&lt;/p>&lt;p>This knowledge has useful meaning, but it does not provide for, in and of itself, an integration such as inferring further knowledge. Also, knowledge is deterministic.&lt;/p>&lt;h5 id="examples-2">Examples&lt;/h5>&lt;ul>&lt;li>Knowing how Spring Boot works, and thus knowing how to operate Spring Boot.&lt;/li>&lt;li>Knowing how an electric hob works, what to press to start it, how long to keep it on to make it reach a certain temperature.&lt;/li>&lt;li>Knowing which underground lines are involved in a journey, where to take the train, where to change, where to hop off.&lt;/li>&lt;/ul>&lt;h4 id="understanding">Understanding&lt;/h4>&lt;p>Understanding answers &amp;ldquo;why&amp;rdquo; questions. Understanding is the process by which people can take knowledge and synthesize new knowledge based on the previously held knowledge. While knowledge copes well with the deterministic, understanding can cope with interpolation and probability.&lt;/p>&lt;p>The difference between understanding and knowledge is the difference between &amp;ldquo;learning&amp;rdquo; and &amp;ldquo;memorizing&amp;rdquo;. People who understand something can infer new knowledge and adapt it to new situations.&lt;/p>&lt;p>Understanding is intrinsically systemic, as you can never understand why a system works they way it works by analysis, by taking the system apart. The reason a system works the way it works does not reside in the system itself, but in its containing system. You cannot, as an example, understand why a university works the way it works by analyzing how its various parts operate. You need to understand the education system that university is part of to even attempt an answer to this question.&lt;/p>&lt;p>Understanding allows to design a system, rather than merely controlling it and operating it.&lt;/p>&lt;h5 id="examples-3">Examples&lt;/h5>&lt;ul>&lt;li>Knowing why Spring Boot works the way it works, by understanding the JVM ecosystem, its history, Dependency Injection, how companies that hire Java developers operate, which technologies are most common in a JVM environment, Cloud providers, etc.&lt;/li>&lt;li>Understanding the web ecosystem, and thus being able to infer how a framework new to you e.g. Vue.js works, based on previous knowledge e.g. React.js, Ajax, JSON, jQuery, JavaScript, etc.&lt;/li>&lt;/ul>&lt;h4 id="wisdom">Wisdom&lt;/h4>&lt;p>While the previous levels help you do the thing right, wisdom is about doing the right thing. Wisdom is therefore the process by which we also discern, or judge, between right and wrong, good and bad.&lt;/p>&lt;p>Wisdom allows to design and create effective systems, which produce the desired outcomes.&lt;/p>&lt;p>The wise person must not only possess great understanding and wide appropriate knowledge, but start from the desired outcomes and walk back to the approach, the decisions, and the instruments, using their knowledge and understanding as selection criteria throughout.&lt;/p>&lt;p>Wisdom is developing mental models about what&amp;rsquo;s desirable, what works, what doesn&amp;rsquo;t, and why, about making predictions on reality based on these models, and about discarding or evolving the models when reality proves them wrong.&lt;/p>&lt;h5 id="examples-4">Examples&lt;/h5>&lt;ul>&lt;li>Achieving effective staffing for a company, given that company&amp;rsquo;s size, goals, strategy, industry, and location, but also given the market in the company&amp;rsquo;s country, the belief system of the executives, the budget, the laws and regulations in place, etc.&lt;/li>&lt;li>Raising children, governing a country, choosing a life-long partner, etc.&lt;/li>&lt;li>A principled approach to sustainable software product development.&lt;/li>&lt;/ul>&lt;h3 id="so-what-how-does-this-help-me-build-a-company-and-better-products">So what? How does this help me build a company and better products?&lt;/h3>&lt;p>Why is important to keep in mind these various categories of what&amp;rsquo;s knowable? Because it&amp;rsquo;s easy to expect unachievable outcomes or focus on the wrong things otherwise.&lt;/p>&lt;p>What&amp;rsquo;s your approach with data and data engineering? Do you use data as a drunk uses a lamppost: for support rather than illumination?&lt;/p>&lt;p>What do you expect from your employees? Knowledge? Or wisdom and understanding? Do you interview accordingly?&lt;/p>&lt;h4 id="some-examples-of-problematic-behaviours">Some examples of problematic behaviours&lt;/h4>&lt;ul>&lt;li>A &amp;ldquo;data-driven&amp;rdquo; strategy. Data asks no questions, but it might give answers to a question when organised and processed correctly. Do you know which questions you&amp;rsquo;re asking your data? How do you know those are the right questions? Data and information are for feedback and model validation only.&lt;/li>&lt;li>Do you ask information questions when you interview candidates e.g. &amp;ldquo;What is the name of the property that does this in Spring Boot&amp;rdquo;? Or do you try and gauge their understanding and wisdom? Why?&lt;/li>&lt;li>Do you understand the limitations of domain knowledge and of knowing an industry? What is this useful for? What can it not ever be enough for?&lt;/li>&lt;li>Do you understand the difference between efficiency and effectiveness? About doing the thing right and doing the right thing? Is your company acting accordingly in this sense?&lt;/li>&lt;li>Do you design your approaches based on the desired end results? Do you walk back from the outcomes to the approaches, the teams, the ways of working, the technologies, etc.? Or do you think doing the thing right will be enough to take you there?&lt;/li>&lt;/ul></description></item><item><title>Lifers and politicians</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-29-lifers-and-politicians/</link><pubDate>Thu, 29 Jun 2023 15:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-29-lifers-and-politicians/</guid><description>&lt;p>I&amp;rsquo;m sure you have encountered plenty of toxic people at work. Let&amp;rsquo;s look at some of the most dangerous kinds: lifers, and politicians.&lt;/p>&lt;p>Toxic employees come in all shapes and sorts. Some are relatively innocuous, while others can kill a company if not addressed. The more obvious it is that somebody is toxic at work, the less dangerous they usually are.&lt;/p>&lt;p>Today I&amp;rsquo;m going to focus on 2 specific types of toxic employees: lifers, and politicians.&lt;/p>&lt;p>Both can be very dangerous and problematic, especially when they manage to rise near the top in an organization. Also, both can be found in any company, and in any age band.&lt;/p>&lt;p>Their behaviour and motivations are very different though.&lt;/p>&lt;h3 id="lifers">Lifers&lt;/h3>&lt;p>Lifers stopped learning a long time ago, stopped caring, and became part of the furniture in their company. The status quo is comfortable to them, and so they&amp;rsquo;ll defend it from any attempt to change things.&lt;/p>&lt;p>Typically, somebody becomes a lifer in a company where mediocrity is tolerated but mistakes carry social or career danger. With no incentives to learn, and no options, fossilizing is inevitable.&lt;/p>&lt;p>They usually end up convincing themselves that the little they know is the only acceptable state, and that everything else is wrong/idealistic/academic/impossible.&lt;/p>&lt;p>Like zombies, they&amp;rsquo;re only dangerous when in numbers.&lt;/p>&lt;p>When they are prevalent in a company:&lt;/p>&lt;ul>&lt;li>There&amp;rsquo;s a lot of attrition among new-joiners. People smash the interview and enthusiastically join only to get depressed and leave within a few months.&lt;/li>&lt;li>Nothing changes ever, except for how things are called. You see a bunch of system admins becoming an &amp;ldquo;SRE team&amp;rdquo;, then a &amp;ldquo;DevOps team&amp;rdquo;, and a &amp;ldquo;Platform Engineering team&amp;rdquo;, without any change whatsoever.&lt;/li>&lt;li>Companies fall behind in terms of technologies, stuck using Java 8, archaic frameworks, and vintage data formats.&lt;/li>&lt;/ul>&lt;h4 id="clues">Clues&lt;/h4>&lt;ul>&lt;li>Saying there&amp;rsquo;s no reason to ever deviate from their current tools and processes. NoSQL, event-driven architecture, Cloud, streaming, reactive programming, mob programming, trunk-based development, all fantasy.&lt;/li>&lt;li>Not reading, watching videos, or working on side projects.&lt;/li>&lt;li>Claiming that things other companies use successfully would never work in their company.&lt;/li>&lt;li>Discrediting new joiners by saying they have no experience running the company systems in production.&lt;/li>&lt;/ul>&lt;h4 id="scorecard">Scorecard&lt;/h4>&lt;ul>&lt;li>Self-awareness: low (they can&amp;rsquo;t cope with the idea that they&amp;rsquo;re obsolete, even if deep down they suspect it)&lt;/li>&lt;li>Competence: very low (stopped learning a long time ago, stopped caring as well)&lt;/li>&lt;li>Maliciousness: low - (mostly defending the image they have of themselves)&lt;/li>&lt;/ul>&lt;h3 id="politicians">Politicians&lt;/h3>&lt;p>Politicians worm their way in and up organizations through intentional deception. They&amp;rsquo;re usually quick to identify people who are receptive to their games, and then spend most of their time manipulating them to improve their own status and power.&lt;/p>&lt;p>These people know what they&amp;rsquo;re doing, and deliberately discredit people, steal recognition, cast blame, undermine others, lie, and bribe.&lt;/p>&lt;p>When they join a company:&lt;/p>&lt;ul>&lt;li>The best people leave, as they typically become victims and get ousted, or they have options and decide they had enough.&lt;/li>&lt;li>Big projects start failing, only for the politicians to deflect the responsibility towards the people they like the least.&lt;/li>&lt;li>More employees start becoming politicians, as the successes of these toxic people incentivize bad behaviour in others.&lt;/li>&lt;/ul>&lt;h4 id="clues-1">Clues&lt;/h4>&lt;ul>&lt;li>Continuously saying they&amp;rsquo;re not political.&lt;/li>&lt;li>Kissing up and kicking down.&lt;/li>&lt;li>Having back-to-back calendars and spending most days in meetings they create themselves.&lt;/li>&lt;li>Spending more time with their managers than with their peers.&lt;/li>&lt;li>Posturing by asking or saying meaningless things that sound reasonable:&lt;ul>&lt;li>Do you have any data to sustain this?&lt;/li>&lt;li>Do we have the agency to make this decision?&lt;/li>&lt;li>We need more accountability and ownership.&lt;/li>&lt;li>We need to collaborate more.&lt;/li>&lt;li>We need Scrum/Kanban/ShapeUp/SAFe/LESS.&lt;/li>&lt;li>We need more diversity.&lt;/li>&lt;li>We need commitments/deadlines/estimates/velocity.&lt;/li>&lt;/ul>&lt;/li>&lt;li>Hiring tons of unnecessary people with their area of control.&lt;/li>&lt;/ul>&lt;h4 id="scorecard-1">Scorecard&lt;/h4>&lt;ul>&lt;li>Self-awareness: very high (they know themselves and their game)&lt;/li>&lt;li>Competence: very low (gave up on the idea of achieving things through competence)&lt;/li>&lt;li>Maliciousness: very high - (willing to harm others and their company to get ahead)&lt;/li>&lt;/ul></description></item><item><title>When it comes to models it's not about truth</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-29-models/</link><pubDate>Thu, 29 Jun 2023 10:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-29-models/</guid><description>&lt;p>Understanding models and lenses, their applicability, and their limitations.&lt;/p>&lt;h3 id="the-philosophical-bits">The philosophical bits&lt;/h3>&lt;p>Life is about luck and decision-making. Luck has tremendous impact, but we don&amp;rsquo;t control it, which leaves us with decisions.&lt;/p>&lt;p>A good decision is a decision that produces good outcomes, according to the decision-maker. Given the situation, the problem, and the constraints, your job is to make good decisions to produce good outcomes. So why is it so hard? Because of complexity, a long time horizon, and limited information.&lt;/p>&lt;p>Reality is complex: everything is interconnected and interdependent, systems are recursively part of larger systems. Tiny changes can have unexpected and non-linear effects, which might manifest themselves after a long time. Throw in the fact that we humans can&amp;rsquo;t cope well with delays in feedback, and that we pretty much never have complete information about the state, the problem, the options, and the constraints, and it&amp;rsquo;s no wonder that consistently making good decisions is so hard.&lt;/p>&lt;p>Which is unfortunate news, because consistently making good decisions is what allows to achieve great outcomes, leaving aside luck. One bad decision can put us in an unsavoury position, and could make it hard or impossible to recover.&lt;/p>&lt;p>Mind that doing the right thing is way more important than doing the thing right. Identifying the best course of action triumphs over executing well a wrong approach. If you want to reach a destination, direction matters way more than speed. Indeed, doing the wrong thing righter makes you wronger, like Ackoff used to say and teach. If you&amp;rsquo;re moving in the wrong direction, the faster you go the further you&amp;rsquo;re getting from where you wanted to be.&lt;/p>&lt;p>So how do we make decisions? Well, we end up doing what we believe is best. This is never objective, and objectiveness and truth don&amp;rsquo;t even matter in this sense, what matters is making the right decision. Our belief system, which is our collective set of mental models, is what determines how we see a problem and what course of action we deem the most appropriate, given the circumstances.&lt;/p>&lt;p>2 people with a different way of thinking and seeing the world would make different decisions if put in the same situation.&lt;/p>&lt;p>Some people think that maths, physics, medicine and so on are true and universal, but this isn&amp;rsquo;t the case. They attempt to model and explain reality and, like every model, they omit details and have local validity. And yet they are tremendously useful: remember, the point isn&amp;rsquo;t being true, but allowing us to make good decisions.&lt;/p>&lt;p>So in order to make good decisions, we need to leverage useful mental models, given the problem and the circumstances. This consists of 2 parts:&lt;/p>&lt;ol>&lt;li>Forming a wide range of mental models.&lt;/li>&lt;li>Adopting the right lens and mental models for the problem and the circumstances.&lt;/li>&lt;/ol>&lt;p>Like building a toolbox and choosing the right tool for the job at hand.&lt;/p>&lt;h3 id="the-implications-for-software-product-development">The implications for software product development&lt;/h3>&lt;p>So what then? Why all this abstract reasoning? Most companies employ plenty of educated, hard-working, smart, and well-meaning people, and yet often they fail to achieve the outcomes they desire.&lt;/p>&lt;p>And I argue that this is due to ineffective mental models at play, which inevitably lead to bad outcomes. In terms of what kinds of mental models I&amp;rsquo;m talking about, I categorise them as follows:&lt;/p>&lt;ol>&lt;li>Organizational: nature and purpose of a company, interdependencies between processes, people, and teams, incentives, collaboration.&lt;/li>&lt;li>Context-specific: factory vs lab metaphor, business model, quantity vs quality, delivery vs development.&lt;/li>&lt;/ol>&lt;p>What has your experience been? Does any of this resonate? have you ever seen a company putting a lot of effort into something without achieving much?&lt;/p></description></item><item><title>Why?</title><link>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-28-why/</link><pubDate>Wed, 28 Jun 2023 15:00:00 +0000</pubDate><guid>https://sollecitom.github.io/software-product-development-blog/posts/2023/2023-06-28-why/</guid><description>&lt;p>I can hear the &lt;em>what&amp;rsquo;s in it for me?&lt;/em> in your head.&lt;/p>&lt;p>I hope you&amp;rsquo;ll find some of my thoughts useful, and this is more likely if you&amp;rsquo;re a startup founder, a software engineer, a product manager, a marketer, a project manager or any other software product development professional.&lt;/p>&lt;p>And what about me? Well, I&amp;rsquo;ve always found that writing things down makes them clearer in my head, and I might occasionally dig up an old article when discussing things with friends or at work.&lt;/p></description></item></channel></rss>