<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Stefan Ellersdorfer on Medium]]></title>
        <description><![CDATA[Stories by Stefan Ellersdorfer on Medium]]></description>
        <link>https://medium.com/@steell0815?source=rss-d51fc814c089------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*e3Mlf_k_RgPFzIlyhbAEuA.png</url>
            <title>Stories by Stefan Ellersdorfer on Medium</title>
            <link>https://medium.com/@steell0815?source=rss-d51fc814c089------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 12:01:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@steell0815/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Obvious Saturday: How Much More Secure Does a Certificate Make You?]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-how-much-more-secure-does-a-certificate-make-you-a5e8df95fdc8?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/a5e8df95fdc8</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[obvious]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 07:01:02 GMT</pubDate>
            <atom:updated>2026-11-13T07:01:02.226Z</atom:updated>
            <content:encoded><![CDATA[<p>This week, I had the opportunity to speak about secure software engineering at a cybersecurity event.</p><p>Naturally, many of the other speakers approached security from an operational perspective.</p><p>Firewalls. Endpoint protection. Identity management. Malware scanners. Content filters. Vulnerability scanners. Intrusion detection and prevention. SIEM. 24/7/365 Security Operations Centers.</p><p>And much more.</p><p>An impressive arsenal of technologies, processes, and organizational measures to protect our increasingly interconnected systems.</p><p>All of them have their place.</p><figure><img alt="Me, sitting on a desk and reiterating notes on an obvious saturday article" src="https://cdn-images-1.medium.com/max/1024/1*euDmOQ9Kz7BDWJxHKiokCw.jpeg" /><figcaption>How much more secure?</figcaption></figure><p>But I started somewhere else. At the point where security considerations are frequently missing, or at least not systematically driven by examples.</p><p>At the very beginning of software engineering. Those who know me probably anticipated my answer: Behavior-Driven Development.</p><blockquote>Again?</blockquote><blockquote>Yes. Again.</blockquote><p>Because BDD, particularly example-driven acceptance testing, offers something rather fundamental.</p><p><strong>The opportunity to turn expectations about system behavior into executable evidence.</strong></p><p>These expectations are not limited to business functionality: Authentication. Authorization. Business rule compliance. Performance characteristics. Security constraints. Even certain resilience and operational requirements.</p><p>We can describe expected behavior using a domain-specific language, express concrete examples, and immediately verify whether the implementation satisfies them.</p><p>Not every security property can be exhaustively tested. But an astonishing number of expectations that otherwise remain buried in documents, tickets, policies, and assumptions can become executable.</p><p>And that’s where things become interesting. Because this changes <strong>when</strong> we think about security. Instead of implementing functionality and subsequently asking whether it is secure, we consider security alongside business behavior from the very beginning.</p><p>Not as a separate analysis phase, but as an upfront design exercise followed by implementation and testing. And certainly not as another waterfall disguised as agile.</p><p>Rather, as one continuous engineering activity:</p><ul><li>Describe an expected behavior through examples. Implement it. Verify it immediately. Learn from the feedback. Repeat.</li><li><strong>Continuous Integration</strong> provides the immediate feedback.</li><li><strong>Continuous Delivery</strong> keeps the system in a releasable state.</li><li>And where the environment permits, <strong>Continuous Deployment</strong> closes the loop even further.</li></ul><figure><img alt="Closed feedback cycle through continuous integration, delivery and deployment" src="https://cdn-images-1.medium.com/max/1024/1*4CLjXTIFuJJV4MqCVUWuFQ.png" /><figcaption>Closed feedback cycle through continuous integration, delivery and deployment</figcaption></figure><p>Because ultimately, there is one reality that matters:</p><p><strong>How the system actually behaves when real people and other systems interact with it.</strong></p><p>Now, let’s return to the operational security arsenal.</p><p>What happens when software engineering continues to treat security as an afterthought, while we generate increasingly large amounts of code, perhaps accelerated by AI, without equally accelerating our ability to verify its behavior?</p><p>When we build systems faster than we can establish confidence in them?</p><p>We <strong>increase the attack surface</strong> and potentially <strong>introduce vulnerabilities</strong> <strong>faster </strong>than our <strong>operational countermeasures can identify</strong>, contain, and remediate them.</p><p>More scanners. More filters. More alerts. More dashboards. More people watching dashboards.</p><p>And eventually, we risk building an extraordinarily sophisticated infrastructure to compensate for an engineering system that has never learned to establish security in the first place.</p><p><strong>As if we had already accepted losing control.</strong></p><p>Which brings me back to certificates: such can provide evidence that an organization has established certain practices, controls, and management processes. That matters. But it does not, by itself, establish that the software we release today behaves securely.</p><p>Nor does a successful penetration test yesterday guarantee that tomorrow’s changes won’t introduce new vulnerabilities.</p><p>Security is not a state we achieve once and subsequently defend. It is a property we must continuously work to establish, verify, and preserve.</p><p>And this is where I believe software engineering has an extraordinary opportunity — to reduce the burden we place on operational security.</p><p><strong>What if our engineering systems produced evidence of secure behavior with every single change?</strong></p><p>Not merely evidence that we followed a process, and not merely evidence that we passed an audit, but executable, continuously renewed evidence that the system satisfies explicitly defined expectations.</p><p>The most interesting question is not how much more secure a certificate makes us, but: <strong>how much of our security can we actually demonstrate, right now?</strong></p><p>Think about it. And about the choice that, once again, I may have made obvious to you.</p><p><strong>Here’s a simple experiment for your team.</strong></p><p>Pick one security requirement in your current system. Authentication, authorization, data access, or something similarly fundamental.</p><p>Explore:</p><ol><li>Can your team demonstrate, through an executable test, that the requirement is satisfied?</li><li>Can you run that verification automatically after every change?</li><li>And would you know immediately if the behavior changed?</li></ol><p>If the answer to any of these questions is <strong>no</strong>, you’ve identified an opportunity to improve your engineering system — not by introducing another security policy, another dashboard, or another certificate, but by turning a security expectation into continuously executable evidence.</p><p><strong>One requirement. One example. One automated verification.</strong></p><p>And then another.</p><p>That’s how security becomes part of everyday software engineering rather than something we inspect afterwards.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a5e8df95fdc8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday: Is There Anything Obvious?]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-is-there-anything-obvious-96b5d8b23035?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/96b5d8b23035</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[obvious]]></category>
            <category><![CDATA[software-craftsmanship]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 06:32:24 GMT</pubDate>
            <atom:updated>2026-10-31T06:32:24.463Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Is knowledge a sticky matter?</em></p><p>Recently, I began writing about obvious topics.</p><p>Obvious to me, at least.</p><p>But deep inside, I know that hardly anything really is obvious.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9CydjdZk5lhkgQf5_cUn1Q.jpeg" /><figcaption>Is There Anything Obvious?</figcaption></figure><p>Nearly everything is contextual. What seems perfectly clear to one person may be entirely unfamiliar, or mean something completely different, to another.</p><p>And that’s fine.</p><p>But there are some ideas and practices whose original definitions were remarkably precise. Not arbitrary. Not merely a matter of interpretation. They emerged from particular problems, were carefully articulated, and came with rather specific intentions.</p><p><strong>What is Continuous Integration?</strong></p><p><strong>What is Refactoring?</strong></p><p><strong>What is Test-Driven Development?</strong></p><p>Why do I consider this important?</p><p>Because apparently, knowledge isn’t particularly sticky.</p><p>An idea is invented, discovered, or advocated for the first time. Some of these ideas become a blast. People get excited. They understand the idea, adopt it, embody it, teach it to others.</p><p>And then something curious happens: as the idea spreads, its original meaning slowly begins to deteriorate. Interpretations replace definitions. Practices become rituals. Myths replace experience. And eventually, the terminology survives while the substance nearly disappears.</p><p>Consider a few examples.</p><p><strong>Continuous Integration</strong> — originally the practice of integrating everyone’s work frequently into a shared, working codebase — becomes a pipeline script. Sometimes running on branches that remain isolated for days or weeks. The automation survives, while the integration disappears.</p><p><strong>Refactoring</strong> — originally the disciplined practice of improving the internal structure of existing code without changing its observable behavior — becomes a huge, one-off effort. A project requiring budget, planning, management approval, and often ending in dramatic failure.</p><p><strong>Test-Driven Development</strong> — originally a disciplined development practice in which tests guide design through short feedback cycles — becomes testing after the fact. Often with tests tightly coupled to implementation details. Sometimes TDD is simply understood as another name for unit testing.</p><p>The names remain, but the practices don’t. And what troubles me is not merely the loss of precision — it is what happens next:</p><p>Organizations start planning, budgeting, managing, measuring, and optimizing these misconceptions.</p><p>They invest considerable effort into doing something they believe represents an established engineering practice, without ever experiencing the benefits that made the original idea valuable.</p><p><strong>A planned, budgeted, project-managed misconception.</strong></p><p>And when it doesn’t work?</p><p>Well, obviously, the practice doesn’t work.</p><p>Refactoring is too expensive. Continuous Integration is complicated. TDD slows developers down.</p><p>Except that perhaps none of the original practices was actually attempted in the first place, which brings me back to the question of what is obvious.</p><p>My learning is, that knowledge isn’t something we can simply transfer, document, or preserve by giving it a name.</p><p>It rather needs to be continuously exercised, challenged, rediscovered, and taught in order to become knowledge.</p><p>Not because the original thinkers were necessarily right about everything. But because understanding <em>why</em> an idea emerged, <em>which problem</em> it was meant to solve, and <em>which conditions</em> make it useful is fundamentally different from merely repeating its name.</p><p>And that, to me, is one of the greatest responsibilities we have as senior engineers.</p><p>Not to guard terminology like some exclusive guild protecting its secrets.</p><p>Not to insist that everything was better in the old days.</p><p>But to help others discover the reasoning behind the practices. To make experience accessible. To challenge misconceptions, including our own.</p><p><strong>To teach.</strong></p><p>Which I increasingly consider one of the most important things a senior engineer can do.</p><p>And that’s one reason I keep investing effort into <em>Obvious Saturday</em>.</p><p>To reveal opportunities hidden behind apparently familiar ideas. To revisit practices whose origins have nearly been forgotten. To question what we think we know.</p><p>And occasionally, to discover that something I considered obvious wasn’t quite right after all.</p><p>Because perhaps the most obvious thing is that <strong>nothing remains obvious unless we keep learning.</strong></p><p>Appreciate you reading. And if you read this far:</p><h3>Here’s something you can try with your team.</h3><p><em>An experiment in making the obvious visible</em></p><p>Pick one engineering principle or practice you regularly apply. Or, more precisely, one you believe you apply.</p><p>Continuous Integration. Refactoring. Test-Driven Development. Whatever seems obvious to everyone.</p><p>Then gather your team and ask a simple question:</p><p><strong>“What is is and how exactly do we practice this in our team?”</strong></p><p>Let everyone describe their understanding and their actual experience. Not how the practice should work, but how it does work in your particular environment.</p><p>And here are some important details:</p><ul><li>Seniors speak last: let the least experienced or least outspoken team members speak first. Otherwise, the first confident explanation may unintentionally become the accepted truth, and everyone else merely confirms it.</li><li>Don’t correct anyone. Don’t debate. Don’t rush toward consensus.</li><li>Write down every perspective, including contradictions.</li><li>Only after everyone’s understanding is visible, return to the source.</li><li>Look up the original definition. I strongly recommend books, preferably the original works that introduced or carefully described the practice.</li><li>Read the definition out loud. Together.</li><li>Go through the examples. Discuss the reasoning. Understand which problem the practice was originally intended to solve, and why its particular mechanisms matter.</li><li>Then compare that understanding with what you’ve collected.</li><li>Where do you agree? Where have you adapted the practice deliberately? Where have you unknowingly replaced something essential with something merely familiar?</li><li>Are you actually achieving the benefits the practice was designed to provide?</li></ul><p>You might discover that your team understands the practice perfectly well.</p><p>You might discover a legitimate and valuable adaptation.</p><p>Or you might discover that you’ve spent years practicing something entirely different while confidently calling it by the original name.</p><p>None of these outcomes is a failure, the point here is not to prove somebody wrong.</p><p>The point is to make understanding a collective achievement rather than an assumption.</p><p>And perhaps that’s what teaching should look like.</p><p>Not seniors explaining what everyone else ought to know.</p><p>But creating the conditions in which everyone, including the seniors, has an opportunity to learn.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=96b5d8b23035" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday — How to Do Software?]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-how-to-do-software-50ad785078f6?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/50ad785078f6</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-craftmanship]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[tdd]]></category>
            <category><![CDATA[continuous-delivery]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 24 Oct 2026 06:01:01 GMT</pubDate>
            <atom:updated>2026-10-24T06:01:01.759Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Don’t panic. We finally have more time to do the engineering.</em></p><p>AI is everywhere these days. And I mean <em>literally</em> everywhere. In literature, columns, news, conferences, LinkedIn posts, and probably somewhere on your local machine. Everywhere. Every now and then, I find myself triggered by yet another remarkably confident statement about how software is going to be developed from now on.</p><p>The new way. The only way. The inevitable future. Software engineers will become obsolete. AI will do everything. Writing code is no longer a skill worth cultivating. Just describe what you want, let the machines do the work, and move on.</p><p>Well.</p><p><strong>Don’t panic.</strong></p><p>I see things rather differently.</p><p>In fact, I believe software engineering is becoming more important than ever. And, perhaps for the first time, we have an extraordinary opportunity to practice it properly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0N0N9YJSbj93LIKLTfb_lw.jpeg" /><figcaption>How to Do Software</figcaption></figure><h3>The iceberg we keep ignoring</h3><p>For decades, much of our industry has treated software development primarily as a code creation activity.</p><p>We write code… …</p><p>… … We review code… …</p><p>… … … … We test code… …</p><p>… … … … … … We deliver code… … and start again.</p><p>In countless projects, quality assurance has largely consisted of human code reviews followed by manual testing. Sometimes a few automated tests. Sometimes a substantial testing phase. Sometimes a heroic stabilization effort shortly before release.</p><p>And somehow, we called that software engineering, when that feels much more like translating then engineering. But creating software is only the visible tip of an iceberg. Beneath the surface lie architecture, coupling, dependencies, maintainability, testability, security, and the economics of change.</p><p>And that’s only the first iceberg. There is another one called <strong>maintenance</strong>. And another one called <strong>operations</strong>. As software rarely exists for its own sake, there are even more icebergs: the businesses, processes, regulations, human activities, and economic systems that our software is supposed to support. Some of those are considerably larger than the software itself. We engineers don’t always understand them particularly well. And neither does AI, merely because it can generate a plausible implementation.</p><h3>Now imagine accelerating the wrong thing</h3><p>Suppose we haven’t substantially improved our engineering capabilities and our architecture remains fragile and prematurely defined. Our tests unreliable, our feedback loops painfully slow if existing at all, and our understanding of the desired outcomes incomplete. Suppose we still rely on human reviewers to catch issues and manual testers to discover whether our assumptions were wrong.</p><p>Now introduce AI.</p><p>Suddenly, we can create code much faster. Much, much faster.</p><p>Wonderful.</p><p>Except that we have accelerated the production of something whose consequences we still cannot reliably evaluate: More code. More dependencies. More interactions. More changes. More potential vulnerabilities. More things to operate, understand, and eventually maintain.</p><p><strong>If code creation accelerates while engineering capability remains unchanged, we are becoming LESS productive.</strong></p><p>We are simply approaching the iceberg faster.</p><p>And, sorry to say, without engineering, we will eventually hit it.</p><p>Hard.</p><h3>What if we used AI for something more interesting?</h3><p>I am a continuous integration and continuous delivery guy, hence I love short feedback loops. I love trunk-based development, test-driven development, and BDD-style acceptance testing. Not because I particularly enjoy writing tests, but because I want to know whether the system still behaves as intended.</p><p>I want executable evidence that our assumptions hold — or immediate feedback when they don’t. Of course, tests cannot prove the absence of defects. Nor can they establish that our understanding of the business was correct in the first place. But they can make explicit expectations repeatably verifiable. And that is considerably more useful than somebody saying, “Looks good to me.”</p><p>Now, with AI at hand, something interesting happens: I can spend less time on the mechanics of producing code and more time strengthening the engineering system around it. Thereby elaborate acceptance tests that express meaningful business outcomes. I can improve the architecture and systematically reduce harmful coupling — even drop parts of existing solutions and replace them with better options. I can introduce mutation testing to challenge whether my tests are actually meaningful. Also I extend vulnerability scanning, dependency analysis, security testing, and automated penetration testing. Establish deterministic checks for formatting, linting, architecture constraints, and static analysis — some executed before a commit, others continuously in the integration pipeline. I improve observability, operational verification, and recovery mechanisms.</p><p>And AI can help me build these capabilities, too.</p><p>Not by replacing deterministic verification with another probabilistic opinion, but by helping me create, maintain, and extend the mechanisms that produce reproducible evidence.</p><p><strong>AI can help us engineer the engineering system.</strong></p><p>That is far more exciting to me than simply generating another thousand lines of code before lunch.</p><h3>The economics of change haven’t disappeared</h3><p>One of the most remarkable misconceptions in the current discussion is that faster code generation somehow makes the economics of software irrelevant. It does not. Code has never been the value itself.</p><p>The value lies in what the software enables people and organizations to accomplish. Code is one means of delivering that value — and simultaneously a liability that must be understood, maintained, secured, and operated.</p><p>AI reduces the cost of creating code dramatically.</p><p>But it does not automatically reduce coupling, eliminate complexity, improve business understanding, or make the resulting system easier to change. In fact, generating more code without understanding its consequences can make all of these problems worse. And the faster we generate it, the faster those consequences accumulate.</p><p><strong>The economics of change are governed by much more than the speed of typing — or generating — code.</strong></p><h3>This is our opportunity</h3><p>For years, we have argued that engineering practices are essential: Automated testing. Continuous integration. Continuous delivery. Refactoring. Security by design. Observability. Fast feedback.</p><p>And for years, we have heard that there simply isn’t enough time.</p><p>We need to deliver features, meet a deadline.</p><p>We will improve the engineering system later.</p><p>Well, now we have a tool that can substantially accelerate parts of implementation.</p><p>What are we going to do with the time it gives us?</p><p>Generate even more code?</p><p>Or finally invest in the capabilities that allow us to deliver valuable, reliable software sustainably?</p><p>I know where I want to spend my time, not because I am nostalgic about programming by hand or because I believe AI cannot produce excellent code. And certainly not because I want to defend my profession against technological progress.</p><p>Quite the opposite.</p><p>I want to use AI to become a better software engineer.</p><p>To understand problems more deeply. To challenge assumptions earlier. To make quality more observable. To build stronger feedback mechanisms. To make systems easier to change, safer to operate, and more useful to the people they serve.</p><p>To do the engineering work we should have been doing all along.</p><p>Only better, more precisely, more comprehensively, even more elegantly.</p><p><strong>AI makes writing code cheaper.</strong></p><p><strong>That gives us an opportunity to make software engineering better.</strong></p><p>Whether we seize that opportunity is not a question of AI capability.</p><p>It is a question of engineering ambition.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=50ad785078f6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday: Trust Is A Resource]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-trust-is-a-resource-a7164e7bc8b0?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/a7164e7bc8b0</guid>
            <category><![CDATA[obvious]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[trust]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 17 Oct 2026 07:39:16 GMT</pubDate>
            <atom:updated>2026-10-17T07:39:16.590Z</atom:updated>
            <content:encoded><![CDATA[<p><em>And micromanagement is one of the most effective ways to waste it.</em></p><p>Trust is not always naturally there.</p><p>In many organizations, leaders do not trust their people. They micromanage instead.</p><p>But what exactly does that mean? And why does it happen?</p><p>As with so many things in software engineering, this coin has two sides.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sMDATL3JhnMk-pjfVI2j4g.jpeg" /></figure><p>On one side, we have engineers who do not necessarily act responsibly. They follow plans, implement specifications, complete tickets, and consider their work done when the requested solution has been delivered.</p><p>On the other side, we have leaders who do not actually invite people to take responsibility. Instead, they assign tasks. Preferably well-prepared ones, with detailed instructions, predetermined solutions, and as little room for interpretation as possible.</p><p>Both sides appear to be doing their jobs.</p><p>And both contribute to a system in which responsibility gradually disappears.</p><h3>The Curious Distribution of Responsibility</h3><p>Consider what happens when an organization wants something built.</p><p>Instead of delegating a problem, together with the authority and budget to decide how to solve it, management delegates a solution.</p><p>Often a remarkably detailed one.</p><p>The architecture has already been decided and technology has been selected too. The user interface has been designed. Eventually the implementation has been broken down into tickets.</p><p>The engineer is expected to execute.</p><p>And because management wants predictability, progress is measured against the plan.</p><p>Did we finish the tickets? Did we implement the specified features? Are we on schedule?</p><p>Interestingly, the very same organizations can be surprisingly <em>laissez-faire </em>when it comes to defining the outcomes they actually want.</p><p><strong>What problem are we trying to solve?</strong></p><p><strong>What would constitute a successful outcome?</strong></p><p><strong>How would we recognize that the solution works as desired?</strong></p><p><strong>Which observable evidence would allow the engineers themselves to determine whether expectations have been met?</strong></p><p>These questions are often answered vaguely, if at all. We prescribe the solution in great detail while leaving the desired outcome remarkably undefined.</p><h3>Imagine Telling a Surgeon How to Operate</h3><p>Imagine you need surgery.</p><p>You consult a surgeon, explain your symptoms, discuss the diagnosis, understand the risks, and agree on the desired outcome.</p><p>And then you hand the surgeon a detailed list of instructions describing precisely where to cut, which instruments to use, and in which order to perform every movement.</p><p>You might even insist on continuous progress reports against your original plan. Absurd?</p><p>Perhaps.</p><p>Or imagine visiting a wonderful fusion restaurant.</p><p>You explain what you enjoy, mention your allergies, discuss your preferences, and then proceed to instruct the chef exactly how to cut the fish and vegetables.</p><p>At what point did you stop buying the expertise of a professional and start buying the execution of your own ideas?</p><p>Of course, surgeons and chefs operate within constraints. They follow professional standards, respect safety requirements, and are accountable for their decisions.</p><p>Professional autonomy is not the absence of constraints. It is the authority to make informed decisions within them. And that is precisely what makes their expertise valuable. Why should software engineering be fundamentally different?</p><h3>The Other Side of the Coin</h3><p>It would be convenient to blame management.</p><p>But that would be too easy. Engineers also contribute to this situation. Some have become remarkably comfortable receiving perfectly prepared tasks. They expect somebody else to understand the problem, make the decisions, resolve the ambiguity, and determine what should be built.</p><p>Their responsibility begins when a ticket enters In Progress and ends when it reaches Done.</p><p>And if the resulting system does not solve the actual problem? Well, then it works as designed. Some engineers even resist being confronted with broader questions.</p><p><strong>Why are we building this?</strong></p><p><strong>Who benefits?</strong></p><p><strong>How will we know it works?</strong></p><p><strong>What alternatives have we considered?</strong></p><p>These questions introduce uncertainty. They demand conversations with people outside the engineering team. They require judgment, and sometimes the uncomfortable admission that we do not know enough yet.</p><p>Following a detailed plan is much easier. But following instructions is not the same as taking responsibility. And expertise that never participates in deciding what ought to be done becomes little more than an expensive execution mechanism.</p><p>Do you wonder whether this might be a source of anxiety in the face of new, highly capable tools — such as LLMs and coding agents?</p><h3>A Self-Reinforcing System</h3><p>Here is where things become particularly interesting.</p><p>Organizations are socio-dynamic systems.</p><p>The behavior of one group influences the behavior of another, and these interactions create reinforcing feedback loops.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x0Kx2K0t9VA1NWP2TXEwIg.png" /><figcaption>A Self-Reinforcing System</figcaption></figure><p>Leaders do not trust engineers to make good decisions, so they make those decisions for them. Engineers are consequently given fewer opportunities to practice judgment, develop ownership, and demonstrate responsibility.</p><p>Some eventually stop trying. Leaders observe this lack of initiative and conclude that their distrust was justified. So they introduce more control.</p><p>More detailed specifications. More approvals. More reporting. More supervision.</p><p>And less room for responsibility. The system produces exactly the behavior it was designed to prevent.</p><p>Not necessarily because anybody intended it. But because the incentives, constraints, and interactions reinforce one another.</p><h3>Trust Is a Resource</h3><p>Trust is not simply a pleasant cultural attribute.</p><p>It is an economic resource. When people trust each other’s professional judgment, they can reduce the effort spent on supervision, approvals, unnecessary coordination, and repeated justification. Decisions can be made closer to the information required to make them and feedback can travel faster. People can spend more of their time solving problems rather than demonstrating that they followed instructions.</p><p>But trust is not blind faith.</p><p>It does not mean abandoning accountability, ignoring risk, or hoping that competent people will somehow produce good results.</p><p>Quite the opposite.</p><p>Trust becomes sustainable when responsibility, decision-making authority, and evidence belong together:</p><ol><li>Give people a problem worth solving.</li><li>Agree on the desired outcomes and the constraints that matter.</li><li>Delegate the authority and resources required to act.</li><li>Establish feedback mechanisms that make progress and results observable.</li><li>And then allow professionals to exercise their expertise.</li></ol><p>In software engineering, this can be remarkably concrete.</p><ul><li>Executable acceptance tests can describe desired behavior without prescribing implementation.</li><li>Automated quality and security controls can establish boundaries without requiring constant supervision.</li><li>Frequent integration and delivery can make progress observable without turning engineers into reporting clerks.</li><li>Direct collaboration with the people who actually use the system can replace layers of interpretation and guesswork.</li></ul><p>These mechanisms do not eliminate the need for trust — they make trust more informed, more resilient, and less dependent on individual heroics.</p><h3>Be Mensch.</h3><p>There is something profoundly human about all of this. We build increasingly sophisticated digital systems, automate processes, accelerate implementation, introduce artificial intelligence, and discover ever more elaborate ways to distribute work.</p><p>Yet we sometimes seem to forget that building useful systems is fundamentally a human activity: it requires understanding and disagreement. It requires the willingness to question assumptions, listen to other perspectives, and occasionally admit that we were wrong.</p><p>And it requires trust.</p><p>Not the kind of trust that says, “Do whatever you want.”</p><p>But the kind that says:</p><ul><li>We have a problem to solve, have agreed on what success looks like.</li><li>We have constraints to respect.</li><li>You bring expertise that I do not have. Let’s figure out how to make this work.</li></ul><p><strong>Leaders</strong> must learn to delegate responsibility rather than merely distribute tasks and <strong>engineers</strong> must learn to accept responsibility rather than merely complete assignments. Both must be willing to participate in the uncomfortable, occasionally messy process of discovering what actually matters.</p><p>Because trust cannot grow where responsibility is systematically denied, and responsibility cannot flourish where nobody is willing to exercise it.</p><p><strong>Be Mensch.</strong></p><p>Be together.</p><p>Discuss. Debate. Disagree. Learn. Conclude.</p><p>And build together.</p><p><strong>After all, it is nothing less than our future that we are building.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a7164e7bc8b0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday: I’m Moving. And So Is My Code.]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-im-moving-and-so-is-my-code-b23a0191c754?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/b23a0191c754</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[obvious]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 10 Oct 2026 19:22:59 GMT</pubDate>
            <atom:updated>2026-10-10T19:22:59.434Z</atom:updated>
            <content:encoded><![CDATA[<p>I’m moving.</p><p>My home, that is.</p><p>And as I go through cupboards, shelves, drawers, and boxes, I keep discovering things I apparently considered worth keeping.</p><p>Years ago… some of them haven’t been used since.</p><p>Others have survived several moves, carefully packed, transported, unpacked, and stored again.</p><p>Just in case.</p><p>And while deciding what deserves a place in my new home, I find myself thinking about software.</p><p>More precisely, about deleting code.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EpPDDpKKvOcLoW5N6rcMdg.jpeg" /><figcaption>I’m moving — and so is my Code</figcaption></figure><h3>What Are We Actually Keeping?</h3><p>We software engineers have a peculiar relationship with code.</p><p>We write it, design and refactor it. Sometimes we’re even proud of it.</p><p>And why shouldn’t we be?</p><p>Good engineering deserves appreciation.</p><p>But there is a fundamental distinction between the craftsmanship involved in creating something and the value of keeping it.</p><p><strong>Code is not the value.</strong></p><p>The value lies in what the code makes possible.</p><p>A business process that becomes more reliable, a task that previously required hours of manual effort. Information that becomes accessible when and where it is needed. A capability that would otherwise be economically or practically impossible.</p><p>Software enables these things.</p><p>Code is merely one of the means by which we achieve them.</p><p>And that changes the economics considerably.</p><h3>Code as a Liability.</h3><p>If the capability is the asset, then what is the code?</p><p><strong>A liability.</strong></p><p>Not necessarily in the strict accounting sense, but certainly in the engineering sense. Code needs to be understood, tested, secured, operated, adapted, and eventually replaced or removed.</p><p>It creates dependencies, introduces complexity and constrains future decisions. And it continues to impose costs long after the excitement of creating it has faded.</p><p>That doesn’t make code bad.</p><p>It makes code something we should be willing to own only when the value it enables justifies the cost.</p><p>Just like the things I’m currently packing.</p><p>A chair I use every day is worth transporting.</p><p>A box containing cables for devices I no longer own, probably not.</p><h3>The Curious Case of Future Use</h3><p>In software engineering, we have a particularly sophisticated version of <strong>just in case</strong>.</p><p>We call it future-proofing.</p><p>We preserve obsolete features because somebody might need them again and maintain configuration options nobody remembers using. Eventually we keep compatibility layers for systems that no longer exist.</p><p>But we don’t stop there.</p><p><strong>We even build entirely new things, sometimes remarkably complex ones, without a demonstrated business case.</strong></p><p>Just in case.</p><p>An abstraction for requirements nobody has expressed, a plugin architecture for extensions nobody has requested.</p><p>A distributed infrastructure for scaling problems we haven’t encountered and a generic framework to accommodate variations that may never exist.</p><p>We invest time, money, and engineering capacity in solving problems that might eventually become real and then we congratulate ourselves on having designed for the future.</p><p>But what exactly have we achieved?</p><p>We have converted a hypothetical future benefit into a very real present liability.</p><p>The code exists. Its dependencies compound and raise complexity. Its maintenance costs exist.</p><p><strong>The value, however, remains hypothetical.</strong></p><p>Of course, anticipating change is part of responsible engineering. Some investments in flexibility are justified by evidence, known constraints, or the cost of making changes later.</p><p>But that’s a business decision, not an architectural virtue in itself. Without such justification, we’re not necessarily future-proofing anything. We’re simply paying today for a future we invented.</p><p>And perhaps the most remarkable part is that we call this good engineering.</p><p>Meanwhile, I’m standing in front of another box of cables I haven’t used in ten years. At least I didn’t buy them in anticipation of devices that might someday be invented.</p><h3>Deletion Is an Engineering Capability</h3><p>Deleting code is not about celebrating fewer lines, nor is it about removing useful behavior… quite the opposite.</p><p>The challenge is to preserve the value while reducing the liability.</p><p>That requires understanding which capabilities actually matter, which behaviors must remain, and which implementation details have become unnecessary. It requires evidence.</p><p>Automated tests. Executable acceptance criteria. Fast feedback. Continuous integration. And the confidence to change a system without losing the capabilities people depend on.</p><p>Sometimes the best outcome of engineering work is not a new component, another abstraction, or a more sophisticated architecture, sometimes it is an entire component that no longer needs to exist.</p><p><strong>Same value. Less liability.</strong></p><p>That’s a rather attractive return on engineering effort.</p><h3>The Things We Fall in Love With</h3><p>Perhaps the hardest part of moving isn’t carrying heavy furniture.</p><p>It’s deciding what to leave behind, because we attach meaning to things. We remember what they cost, why we bought them, how much effort went into acquiring them. And somehow, that history becomes an argument for keeping them.</p><p>Software engineers do the same.</p><p>We remember the difficult problem we solved. The elegant abstraction we invented. The framework we integrated. The architecture we carefully designed. But yesterday’s effort doesn’t establish tomorrow’s value.</p><p><strong>We should fall in love with the value our software creates, not with the code we wrote to create it.</strong></p><p>Once we accept that distinction, deletion as a concept becomes much easier to appreciate. As responsible engineering not destruction.</p><h3>Let’s Move</h3><p>So here I am… surrounded by boxes, wondering why I still own things I haven’t needed in years. Some will come with me, others won’t.</p><p>And perhaps that is a useful way to think about software, too.</p><p>If something creates value, keep it.</p><p>If the same value can be delivered with less code, simplify it.</p><p>If something no longer serves a purpose, remove it.</p><p>And if its only justification is that it <em>might</em> become useful someday, question whether that possibility is worth the cost of keeping it, because moving things isn’t free.</p><p>Neither is maintaining code.</p><p><strong>Code is a liability. Value is the asset.</strong></p><p>Let’s move.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b23a0191c754" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Software is not the Point]]></title>
            <link>https://medium.com/@steell0815/software-is-not-the-point-f8eb97530580?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/f8eb97530580</guid>
            <category><![CDATA[obvious]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 03 Oct 2026 17:26:28 GMT</pubDate>
            <atom:updated>2026-10-03T17:26:28.799Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Some thoughts after GOTO Copenhagen 2026.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*784bC1HXxCZn_5I_5bgZ0Q.jpeg" /><figcaption>Software is not the Point</figcaption></figure><p>Software has no purpose of its own.</p><p>That sounds almost offensively obvious.</p><p>We do not build software because the world desperately needs more software. We rather build it because we want something else to become more effective, more efficient, safer, more reliable, or simply possible.</p><p>An airline does not exist to operate software.</p><p>A hospital does not exist to operate software.</p><p>A bank does not exist to operate software.</p><p>And yet aviation, medicine, finance, manufacturing and almost every other modern industry increasingly depend on digital systems to operate.</p><p>Ideally, software makes those businesses better, annoyingly often, it doesn’t. Sometimes we even turn the relationship around.</p><h3>Humans for AI</h3><p>One recurring theme around AI in software engineering is how we should change the way we work so AI can produce code more effectively. So we restructure tasks for agents, write specifications so the model understands them and optimize repositories for machine consumption. We change workflows because the AI performs better when work is presented in a particular way.</p><p>Some of that is perfectly reasonable.</p><p>But there is a rather important boundary hidden in there:</p><p><strong>AI should be optimized for humans.</strong></p><p><strong>Humans should not be optimized for AI.</strong></p><p>The purpose of introducing AI into software engineering cannot be to make humans better interfaces for code-generating machines. That is not good enough.</p><p><em>How do we organize software development so AI can generate as much code as possible?</em></p><p>This is not an interesting question, but consider this:</p><p><em>How can AI help humans build better systems?</em></p><p>That distinction becomes particularly important because code itself is not the outcome. A functioning business capability is.</p><h3>Regulation</h3><p>We sometimes discuss regulating software engineering as though software were an industry with a single purpose and a reasonably uniform definition of correctness, but software is embedded in almost everything.</p><ul><li>Software flies aircraft.</li><li>Software administers medication.</li><li>Software transfers money.</li><li>Software controls machines.</li><li>Software sells shoes.</li><li>Software shows cat videos.</li></ul><p>These systems should obviously not be governed by precisely the same rules.</p><p>The rules come primarily from the activity the software enables:</p><p>A bank brings financial regulation, risk management and auditability, a medical system brings patient safety, clinical processes and medical-device regulation. Aviation brings another set of constraints, public administration brings another.</p><p>The digital system must adopt these constraints because it participates in that business.</p><p>Trying to define one comprehensive set of rules for “software” therefore seems rather ambitious. It is a little like trying to write a single constitution for all members of the UN Security Council.</p><p>There may be useful common principles.</p><p>But the purpose, risks and constraints remain fundamentally different.</p><p>And this is why digitalization is not simply the act of replacing paper with screens. The digital system becomes part of the operating model of the organization.</p><ul><li>If the process is regulated, the software participates in that regulation.</li><li>If evidence is required, the software must help produce evidence.</li><li>If safety matters, the engineering system must make safety observable.</li><li>If a rule matters, ideally it becomes executable.</li></ul><h3>And then there is engineering</h3><p>Before regulating how software should be built, we would need at least some agreement about what <strong>good software engineering</strong> looks like.</p><p>And surprisingly many organizations do not know (non-exhaustive list):</p><ul><li>Manual tests are considered the quality control.</li><li>A code review is considered a security control.</li><li>A test phase is considered evidence.</li><li>A ticket is considered a specification.</li><li>A deployment pipeline is called Continuous Delivery even though releases happen every few months.</li><li>A bug backlog is accepted as a perfectly ordinary inventory.</li><li>And when AI enters such a system, it does not magically turn it into a mature engineering organization.</li></ul><p>It can simply make the existing system faster, including the bad parts.</p><p>AI can produce more code, analyze more code, can help formulate tests, explore alternatives, explain systems, investigate failures and reduce enormous amounts of mechanical work.</p><p>But none of this tells us what <strong>good</strong> is.</p><p>For that we still need engineering principles (non-exhaustive list):</p><ul><li>Fast feedback</li><li>Executable evidence.</li><li>Automated testing.</li><li>Continuous integration.</li><li>Security controls.</li><li>Observability.</li><li>Small changes.</li><li>The ability to refactor.</li></ul><p>And an engineering system that makes undesirable states difficult to sustain. All these things were useful before generative AI and hence they become more important when producing code becomes dramatically cheaper.</p><h3>The economics of change</h3><p>There is another asymmetry that becomes particularly interesting in the age of AI.</p><p>Speed acts mostly as a factor. Complexity acts like an exponent.</p><p>If AI allows us to produce code twice as fast, we have improved one factor in the economics of software creation.</p><p>But every additional dependency, abstraction, state transition, interface and assumption can interact with what is already there. Complexity does not merely add code. It increases the number of relationships we need to understand when changing the system. And software is changed far more often than it is initially written. So even if generating code became ten times faster, or almost free, these economics do not disappear.</p><p>Quite the opposite: the cheaper code becomes to produce, the easier it becomes to create complexity whose cost will only become visible later.</p><p>Code is cheap. Change is expensive. Complexity determines how expensive.</p><p>The interesting achievement is not producing 10,000 lines of code in an afternoon. It is producing a system that we can still understand, test, operate and safely change three years later. The cost of creating code is dramatically reduced, but the economics of change stay the same.</p><p>The interesting question at this point is not to create more code and more features in the same time, but rather to create five different variants in the same time, compare and weight options. Involve customers and bring the benefit to the human.</p><h3>Perhaps we have the direction backwards</h3><p>After several days of conversations at <strong>GOTO Copenhagen</strong>, I keep returning to the same thought.</p><p>We spend a remarkable amount of energy asking how humans must change for technology.</p><p>How should developers work for AI?</p><p>How should organizations restructure themselves around tools?</p><p>How should software be regulated?</p><p>Perhaps the more useful direction is the opposite.</p><p><strong>AI should adapt to humans.</strong></p><p><strong>Software should adapt to the business it serves.</strong></p><p><strong>Engineering should provide evidence that the resulting system actually does what we need it to do.</strong></p><p>Technology is extraordinarily useful, but it is still the means.</p><p>The human, the organization and the purpose come first.</p><p>Which should be obvious.</p><p>And therefore probably deserves saying.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f8eb97530580" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Conference of the Fearless]]></title>
            <link>https://medium.com/@steell0815/the-conference-of-the-fearless-b7d890825e46?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/b7d890825e46</guid>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Wed, 30 Sep 2026 20:37:21 GMT</pubDate>
            <atom:updated>2026-09-30T20:37:21.982Z</atom:updated>
            <content:encoded><![CDATA[<p><em>In one of the most beautiful cities of the Releasable Kingdom, the fearless gathered once a year.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XSy2tA0cISPglPhEB0kDSQ.jpeg" /><figcaption>Guild of Travelling Orators</figcaption></figure><p>Nobody quite remembered why they were called the fearless.</p><p>Some claimed it was because they travelled great distances to stand before several hundred craftspeople and explain things that had seemed considerably more obvious while preparing the slides. Others believed it was because they willingly performed live demonstrations in front of an audience.</p><p>The Witch considered the second explanation entirely sufficient.</p><p>The gathering was called the <strong>Guild of Travelling Orators</strong>.</p><p>It attracted architects, programmers, masters, apprentices, scholars, merchants, and people whose job titles required surprisingly small letters on their name badges.</p><p>They came to tell stories, and they were magnificent stories.</p><p>One scholar spoke of the Great Dragons of the West, creatures sitting upon such enormous mountains of gold that they had begun investing it in young enterprises whose purpose was remarkably difficult to explain.</p><p>“They are worth twelve billion crowns,” said the Apprentice.</p><p>“What do they make?”</p><p>There was a pause.</p><p>“Mostly presentations explaining why they are worth twelve billion crowns.”</p><p>The Witch nodded.</p><p>“A mature business model.”</p><p>Elsewhere, craftspeople reported their experiences working with Aitch-Two-Oh, the increasingly capable construct that had recently appeared throughout the Kingdom.</p><p>Some called it an assistant, others an agent. Still others called it a colleague.</p><p>One craftsman explained that he spent his days pairing with it.</p><p>“So you work together?” asked the Witch.</p><p>“Yes.”</p><p>“You discuss alternatives?”</p><p>“Sometimes.”</p><p>“You challenge each other’s assumptions?”</p><p>“Well.”</p><p>“You learn from each other?”</p><p>“Certainly. Mostly it learns from everything.”</p><p>“And when it produces nonsense?”</p><p>“I correct it.”</p><p>“And when you produce nonsense?”</p><p>The craftsman hesitated.</p><p>The Witch scratched the cat behind its ears.</p><p>“Perhaps only one of you is pairing.”</p><p>There were also craftspeople who spoke of an astonishingly radical way of working.</p><p>Whenever they changed their work, they brought that change together with everyone else’s almost immediately.</p><p>The Apprentice listened carefully.</p><p>“You mean they don’t keep their work separate for three weeks?”</p><p>“No.”</p><p>“And they don’t create several additional branches so that everyone can work independently?”</p><p>“Apparently not.”</p><p>“And then they don’t finally bring everything together shortly before release?”</p><p>“No.”</p><p>The Apprentice looked concerned.</p><p>“When do they integrate, then?”</p><p>“Continuously.”</p><p>“That sounds suspiciously like Continuous Integration.”</p><p>“Names occasionally outlive their definitions,” said the Witch.</p><p>And then there were the others.</p><p>They were almost disappointing.</p><p>They had no dragons.</p><p>No revolutionary machines.</p><p>No enormous transformation programme.</p><p>No framework with seven differently coloured boxes.</p><p>They talked about behaviour.</p><p>Small changes.</p><p>Pairing.</p><p>Feedback.</p><p>Talking to users.</p><p>Keeping the work releasable.</p><p>Removing unnecessary complexity.</p><p>Bringing work together early.</p><p>Observing what happened.</p><p>And then doing it again.</p><p>They seemed almost like extras in the grand spectacle of technological progress.</p><p>Which, the Apprentice thought, was perhaps a little mean.</p><p>So he asked the Witch.</p><p>“Why are they here?”</p><p>The Witch looked across the hall.</p><p>“Because they do the work.”</p><p>“Everyone here does the work.”</p><p>“Of course.”</p><p>She watched another speaker explain how his team had carefully changed a system, observed what happened, discovered that they had been wrong, and changed it again.</p><p>“No transformation?”</p><p>“No.”</p><p>“No revolution?”</p><p>“No.”</p><p>“No ten-year prediction about the future of our craft?”</p><p>“Apparently not.”</p><p>The Apprentice frowned.</p><p>“That doesn’t sound particularly fearless.”</p><p>The Witch smiled.</p><p>“Stand before a room full of clever people and tell them that your greatest insight is to integrate your work every few minutes.”</p><p>She pointed towards the stage.</p><p>“Tell architects that sometimes the best architecture lies in a decision you haven’t made yet.”</p><p>Another speaker walked past them.</p><p>“Tell programmers that producing more code might not be progress.”</p><p>Another passed by, carrying a laptop covered in stickers.</p><p>“Tell managers that uncertainty cannot be removed by adding more detail to a plan.”</p><p>Then Aitch-Two-Oh rolled past.</p><p>“And tell everyone that having a machine capable of producing answers very quickly makes it more important — not less — to know whether those answers are any good.”</p><p>The Apprentice considered this.</p><p>“That could make people uncomfortable.”</p><p>“Precisely.”</p><p>That evening, the fearless left the conference hall and disappeared into the beautiful streets of the city.</p><p>Some carried new techniques with them.</p><p>Some new questions.</p><p>Some books.</p><p>Some stickers.</p><p>And some carried ideas they had been absolutely certain about that morning, but were now considerably less certain about.</p><p>The Witch considered this an excellent outcome.</p><p>The Apprentice was still thinking about the name.</p><p>“So that’s why they call it the Conference of the Fearless?”</p><p>“Oh, no,” said the Witch.</p><p>“I have no idea why they call it that.”</p><p>She pulled her coat tighter against the northern wind.</p><p>“But bringing together people who are willing to show what they actually do, what they learned from doing it, and occasionally where they were wrong …”</p><p>She looked back at the conference hall.</p><p>“… would be a rather good reason.”</p><p><a href="https://the-effective-software-engineer.com/chronicles/#__tabbed_1_2"><em>Read more.</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b7d890825e46" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday: Professional Helplessness]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-professional-helplessness-1da5ef30a6d7?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/1da5ef30a6d7</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[obvious]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 26 Sep 2026 06:01:01 GMT</pubDate>
            <atom:updated>2026-09-26T06:01:01.757Z</atom:updated>
            <content:encoded><![CDATA[<p>The software industry — and software engineering in particular — is populated by typically bright, sometimes wonderfully geeky professionals usually paid to deal with a continuous stream of challenges.</p><p>Some of those challenges are intellectually fascinating, others are painfully boring.</p><p>But solving problems is, after all, a substantial part of the profession.</p><p>There are even attempts to describe the personalities attracted to software engineering. Depending on which model one chooses, software engineers are frequently characterized as analytical, logical, systematic and rather introverted people.</p><p>People who like to understand how things work and identify inconsistencies.</p><p>People who solve problems.</p><p>Which makes one phenomenon in our industry particularly fascinating: <strong>Professional helplessness.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2jeYd94iGJkdf8bcKM0RbQ.jpeg" /></figure><p>Imagine a software engineer noticing that the build regularly takes forty minutes.</p><p>They know why and even more so, know how to improve it.</p><p>They can probably calculate that fixing it would take less time than the organization collectively looses to it every week.</p><p>And yet:</p><blockquote>Yes, we know. But that’s not part of the current story.</blockquote><p>Or the engineer discovers that deployments regularly fail because an important step is performed manually. They know how to automate it.</p><blockquote>That’s operations.</blockquote><p>A test suite takes two hours and fails randomly.</p><blockquote>We don’t have a ticket for that.</blockquote><p>A dependency has known vulnerabilities.</p><blockquote>Security owns that.</blockquote><p>The architecture makes every small change unnecessarily expensive.</p><blockquote>That’s something we should discuss with the architects.</blockquote><p>Nobody understands an important business rule.</p><blockquote>The specification says…</blockquote><p>And so an interesting transformation takes place.</p><p>The professional problem solver identifies a problem, understands it, sometimes even knows how to solve the problem, and nevertheless learns to continue working around it.</p><p>Not because they are incapable.</p><p>Because they have learned that <strong>capability and authority are two different things</strong>.</p><h3>Learning Not to Act</h3><p>Organizations usually don’t create professional helplessness deliberately.</p><p>Quite the opposite, usually they introduce processes to create clarity, establish responsibilities, make organizations observable and derive measures through which they hope to understand and control the system:</p><ul><li>Responsibilities &amp; Roles.</li><li>Tickets, Backlogs, Budgets, Projects.</li><li>Approval processes.</li><li>Architecture boards.</li><li>Security and Operations departments.</li><li>Product management.</li><li>And measures, KPIs, Dashboards, Velocity, Utilization.</li><li>Ticket counts, Lead times, Bug counts, Story points.</li></ul><p>All of these can serve a perfectly reasonable purpose. We measure systems because we want to understand their internals. We want feedback and need to recognize patterns. We want to learn whether an intervention actually improved something. And sometimes we want control.</p><p>There is nothing inherently wrong with that.</p><p>The interesting part begins when the representation of reality becomes more important than reality itself.</p><p>Because measures are often surprisingly easy to satisfy without improving the system they are supposed to represent.</p><p>If closing tickets is rewarded, tickets get closed.</p><p>If utilization is measured, people become busy.</p><p>If velocity matters, story points become remarkably productive.</p><p>If the number of open bugs matters, bugs can be reclassified, deferred, closed… or moved into a backlog.</p><p>The numbers improve.</p><p>Reality does not necessarily participate.</p><p>And so organizations can become extraordinarily successful at producing what might be called <strong>theatrical truth</strong>: A representation of reality in which the expected numbers are met while the underlying system remains largely unchanged. The dashboard is green, the sprint was successful and the release was delivered. Hence KPI was achieved… but the customer still cannot do what they wanted to do.</p><p>There is a wonderfully precise phrase for this:</p><blockquote>Works as designed.</blockquote><p>Which is not necessarily the same thing as:</p><blockquote>Works as desired.</blockquote><p>The first compares reality with our specification, the second compares reality with the outcome we actually wanted. The difference between those two sentences contains a surprisingly large part of the problem.</p><h3>The Map Becomes the Work</h3><p>This phenomenon does not stop with measurement.</p><p>The ticket represents the work, the backlog represents demand. Velocity represents progress, Bug counts represent quality. Utilization represents productivity. The specification represents the desired system. And increasingly, the Product Manager or Product Owner represents the user.</p><p>Each of these can be useful.</p><p>The problem begins when the representation replaces the thing it was supposed to help us understand. The user is particularly interesting in this regard. Software exists because somebody wants to achieve something with it. Ideally, professionals can observe that person, talk to them, understand their problems, see how they actually work, put something into their hands and learn from what happens next.</p><p>Instead, many organizations construct a chain of representations.</p><p>The user has a problem.<br>→ Someone talks to the user.<br>→ Someone interprets what they heard.<br>→ A Product Manager or Product Owner turns that interpretation into requirements.<br>→ Requirements become backlog items.<br> → Backlog items become stories.<br> → Stories acquire acceptance criteria.<br> → The engineer implements the acceptance criteria.<br> → And eventually the result returns to the user.</p><p>At every step, perfectly reasonable people make perfectly reasonable assumptions. But assumptions compound and what began as observation gradually becomes interpretation.</p><p>Interpretation becomes specification, specification becomes commitment and commitment eventually becomes “Works as designed”.</p><p>The actual user may have disappeared from the process long before the first line of code was written, when the Product Manager or Product Owner has become the organizational representation of the user.</p><p>But representation is not observation.</p><p>And even with the best intentions, that representation necessarily contains assumptions, interpretation and guesswork. Sometimes informed guesswork, sometimes carefully researched and sometimes, unfortunately, rather poorly informed guesswork.</p><p>Yet once encoded into the backlog, guesswork acquires an impressive amount of authority.</p><p>The engineer is no longer primarily asked:</p><blockquote>Does this solve the user’s problem?</blockquote><p>They are asked:</p><blockquote>Does this satisfy the acceptance criteria?</blockquote><p>Those are different questions.</p><p>Although the second one can be an excellent question.</p><h3>Acceptance Criteria and Executable Evidence</h3><p>Acceptance criteria themselves are not the problem, quite the opposite.</p><p>If acceptance criteria describe observable examples of desired behavior and are written as executable tests next to the implementation, they become something remarkably useful: <strong>persistent, executable evidence of our current understanding of what the software is supposed to do.</strong></p><p>They run when the software changes and fail when behavior we explicitly agreed upon no longer holds. Hence they provide feedback to engineers while the work is happening rather than documentation to somebody inspecting the work afterwards. And, importantly, they can be discussed with the people who actually need the behavior before we implement it.</p><p>They do not prove that we wanted the right thing, but they can continuously prove that the thing we built still behaves as we agreed it should.</p><p>That distinction matters.</p><p>Unfortunately, that is surprisingly often not what we call acceptance criteria.</p><p>More commonly, they are a few sentences written into a ticket.</p><p>Sometimes precise, sometimes ambiguous, sometimes merely repeating the requirement. And sometimes reaching almost philosophical levels of optimism.</p><p>I have more than once encountered acceptance criteria resembling:</p><blockquote>Should not break anything.</blockquote><p>A wonderfully ambitious quality goal.</p><p>Unfortunately, not an executable one.</p><p>What exactly is <em>anything</em>?</p><p>What does <em>break</em> mean?</p><p>How would we know?</p><p>Who decides?</p><p>And how long after the implementation are we still observing whether <em>anything</em> has broken?</p><p>Without an executable mechanism answering those questions, “should not break anything” provides neither acceptance nor a criterion.</p><p><strong>It provides hope.</strong></p><p>And this illustrates the larger problem rather nicely.</p><ul><li>The ticket represents the work.</li><li>The Product Owner represents the user.</li><li>The acceptance criteria represent the desired behavior.</li><li>And a checkbox represents evidence that the desired behavior was achieved.</li></ul><p>Each abstraction can be useful.</p><p>But the further we move from observation and executable evidence toward representations of representations, the easier it becomes to satisfy the process without establishing whether the underlying claim is actually true.</p><ul><li>The story is Done.</li><li>The acceptance criteria are checked.</li><li>The implementation works as designed.</li></ul><p>And perhaps somebody, somewhere, will eventually discover whether it also <strong>works as desired</strong>.</p><h3>Managing Problems</h3><p>A bug backlog is another fascinating manifestation of the same phenomenon.</p><p>A defect is evidence that the system does not behave as desired. Put enough defects into a managed backlog, however, and they slowly transform.</p><p>They become prioritized, categorized, measured inventory, perhaps even with an acceptable growth rate.</p><p>The problem has not disappeared.</p><p>But organizationally, it has become manageable.</p><p>And this is an important distinction.</p><p><strong>Managing a problem is not the same thing as solving it.</strong></p><p>Once the problem has an identifier, a category, a priority and an owner, the organization can report on it, which is remarkably comforting.</p><p>The ticket exists, someone owns it.</p><p>The backlog contains it and a dashboard counts it.</p><p>The process works, but the software may not.</p><h3>Reality Does Not Come in Tickets</h3><p>A professional engineer does not encounter reality in conveniently separated tickets. Reality presents systems:</p><ul><li>The slow build affects development.</li><li>The unreliable deployment affects operations.</li><li>The missing executable acceptance test affects our evidence of product behavior.</li><li>The vulnerable dependency affects security.</li><li>The architectural decision affects tomorrow’s development cost.</li><li>The misunderstood user need affects all of them.</li></ul><p>The customer experiences these things as one thing: <strong>the software.</strong></p><p>The organization, however, may experience them as five different responsibilities.</p><p>Each boundary is perfectly reasonable when viewed individually.</p><p>But collectively they create an interesting message: <strong>Do what you have been assigned to do.</strong></p><p>And consequently: <strong>Do not do what you have not been assigned to do.</strong></p><p>That distinction is much more consequential than it first appears.</p><h3>Responsibility Without Authority</h3><p>Work can be delegated.</p><p>Responsibility can be delegated.</p><p>Accountability can be assigned.</p><p>But unless <strong>decision-making authority is delegated alongside the work</strong>, we do not have empowered professionals; we have degraded them to executors.</p><p>Someone else decides what is important, what may be changed, when it may be changed, and how much improvement is permissible.</p><p><strong>The professional is left with the responsibility to execute those decisions.</strong></p><p><strong>This is particularly absurd in software engineering.</strong></p><p>We hire people because of their ability to analyze complex systems, recognize problems, evaluate alternatives, make trade-offs and continuously learn from feedback. Then we carefully construct processes that separate those abilities from the authority to act upon them.</p><p>The engineer sees the slow build, understands its cost and knows how to improve it, but someone else must prioritize the improvement.</p><p>The engineer discovers that the specified feature does not address the user’s actual problem, but someone else owns the requirements.</p><p>The engineer sees an architectural decision increasing the cost of every subsequent change, but someone else owns the architecture.</p><p>The engineer discovers a security problem, but someone else owns security.</p><p>The engineer discovers that “should not break anything” is not actually a testable criterion, but the ticket has already been refined, estimated and committed.</p><p>So the rational professional learns:</p><p><strong>Seeing a problem does not imply having the authority to solve it.</strong></p><p>And eventually:</p><p><strong>Seeing a problem does not even imply that raising it is worth the effort.</strong></p><p>That is where professional helplessness begins. Not with incapable people, rather with capable people repeatedly discovering that decisions about their work are made somewhere else.</p><h3>From Professionals to Executors</h3><p>This becomes particularly interesting when organizations simultaneously ask for <em>ownership</em>.</p><blockquote>Take ownership.</blockquote><blockquote>Be accountable.</blockquote><blockquote>Act proactively.</blockquote><blockquote>Think beyond your ticket.</blockquote><p>These are reasonable requests.</p><p>But ownership cannot extend beyond the authority to make decisions.</p><p>If an engineer is accountable for an outcome but requires permission from four organizational units to change the system producing that outcome, we have not created ownership.</p><p>We have created <strong>responsibility without authority</strong>.</p><p>And people learn remarkably quickly which problems are worth mentioning, which improvements will never be prioritized and which boundaries shouldn’t be crossed.</p><p>They learn that fixing something without a ticket creates more questions than leaving it broken, and they learn what the organization considers reality.</p><p>If the dashboard is green, arguing that the underlying system is unhealthy becomes increasingly difficult, if the specification has been implemented, arguing that the result is wrong becomes somebody else’s concern.</p><p>If the defect is recorded in the bug backlog, the problem is being <em>managed</em>. If the Product Owner accepted the story, apparently the user is satisfied.</p><p>If the acceptance criteria have been checked, apparently the behavior has been verified.</p><p>Eventually they stop asking.</p><p><strong>This is where professional helplessness becomes particularly expensive.</strong></p><p>Because from the outside, nothing dramatic has happened, the engineer still attends meetings, tickets still move, code is still committed. Sprints still end, velocity is still measured and may even look perfectly healthy, the dashboard may be beautifully green and the software may work exactly as designed.</p><p>But something important has disappeared: <strong>The professional no longer feels responsible for improving the system in which the work happens.</strong></p><h3>The Organization Gets What It Optimizes For</h3><p>Organizations sometimes interpret this behavior as a problem with the people.</p><blockquote>They don’t take ownership.</blockquote><blockquote>They aren’t proactive.</blockquote><blockquote>They only work on their tickets.</blockquote><p>So another workshop is organized, perhaps one about empowerment or accountability. Probably a new KPI is introduced.</p><p>But empowerment is not something that can simply be declared.</p><p><strong>Decision-making authority must be delegated alongside the work.</strong></p><ul><li>Measures must remain instruments for learning about the system rather than targets for proving that the system works.</li><li>Product Managers and Product Owners can help us understand users, but they cannot replace contact with them.</li><li>Specifications can communicate understanding, but they cannot replace learning.</li><li>Backlogs can organize work, but they cannot define reality.</li><li>Acceptance criteria can express expected behavior, but prose and checkboxes cannot replace executable evidence.</li><li>Metrics can make aspects of a system visible, but they cannot become the system’s purpose.</li></ul><p>If every meaningful improvement still needs to compete for permission, budget and priority somewhere else, the organizational message remains perfectly clear:</p><p>Don’t improve the system - deliver the assigned work - meet the numbers.</p><p>And intelligent people adapt - professional helplessness is therefore not the absence of professionalism at all. Perhaps it is the perfectly rational behavior of professionals who have learned the actual boundaries of the system they work in.</p><p>We hire them as professional problem solvers.</p><p>We give them complex systems to understand.</p><p>We measure those systems because we want to learn.</p><p>We expect professionals to recognize risks, make trade-offs and take responsibility for outcomes.</p><p>Then we replace users with representations, outcomes with requirements, behavior with prose, evidence with checkboxes, reality with measures and professional judgment with distributed decision-making structures.</p><p>And finally we retain the authority to make the decisions required to change any of it.</p><p>Eventually, professionals adapt to the role we have actually given them:</p><p><strong>executors of somebody else’s decisions.</strong></p><p>And then we wonder why they “don’t take ownership.”</p><p>But ownership without decision-making authority is not ownership.</p><p>It is <strong>accountability for somebody else’s decisions.</strong></p><p>Measurement without the freedom to respond to what we learn is not feedback, just <strong>reporting.</strong></p><p>Acceptance without executable evidence is <strong>hope </strong>not verification.</p><p>And a specification fulfilled without validating the desired outcome is <strong>compliance</strong> not necessarily success.</p><p>Perhaps that is professional helplessness:</p><p><strong>Hiring intelligent people to understand complex problems, surrounding them with increasingly sophisticated instruments to observe those problems.</strong></p><p><strong>And then teaching them that solving what they see is somebody else’s decision.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1da5ef30a6d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Measure of Unexpected Necessities]]></title>
            <link>https://medium.com/@steell0815/the-measure-of-unexpected-necessities-e22221001220?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/e22221001220</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[estimates]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sun, 20 Sep 2026 21:29:52 GMT</pubDate>
            <atom:updated>2026-09-20T21:29:52.461Z</atom:updated>
            <content:encoded><![CDATA[<p><em>There are two prices for building a house in the Releasable Kingdom.</em></p><p><em>The first determines whether you can afford to begin.</em></p><p><em>The second determines whether you could afford to have begun.</em></p><p>Few things were calculated with greater precision than the final cost of work that had been estimated incorrectly.</p><p>Every stone was measured.<br>Every brush was counted.<br>Every hour was recorded.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4uVy-DlTIWBBWCvrCe-5iw.jpeg" /><figcaption>The Measure of Unexpected Necessities</figcaption></figure><p>This precision traditionally began shortly after the citizen had agreed to a considerably smaller number.</p><p>There was, of course, a proper procedure for building a house.</p><p>Before laying a single stone, a prudent citizen would invite the appropriate craftsmen to inspect the site and submit their offers.</p><p>The masons would measure the walls.</p><p>The carpenters would inspect the beams.</p><p>The roofers would count tiles.</p><p>The painters would study the surfaces with the grave concentration of people who knew that paint behaved differently depending on what one attempted to attach it to.</p><p>Each would then prepare a document describing the work to be done, the materials required, and, most importantly, the amount of money for which they were prepared to do it.</p><p>These documents were called *offers*.</p><p>They were regulated by several Royal Acts, three Guild Ordinances and a remarkably thick volume entitled *The Proper Measurement of Works and Materials*, which ensured that citizens could confidently compare them.</p><p>And compare them they did.</p><p>One mason offered to complete the stonework for four thousand crowns.</p><p>Another required four thousand eight hundred.</p><p>A third, apparently unfamiliar with the customs of the kingdom, asked for five thousand six hundred.</p><p>The citizen chose the first.</p><p>This was considered sensible.</p><p>Work began on Monday.</p><p>On Tuesday, the Master Mason discovered that the lower stones required additional preparation.</p><p>On Wednesday, several joints required a different mortar.</p><p>On Thursday, a section of wall had to be strengthened.</p><p>On Friday, additional scaffolding was required to reach a place that had been visible during the original inspection but had apparently become considerably higher since then.</p><p>The citizen watched all this with interest.</p><p>“How much of this was included in the offer?” he eventually asked.</p><p>The Master Mason looked surprised.</p><p>“None of it.”</p><p>“Then why are you doing it?”</p><p>“Because it is necessary.”</p><p>The citizen considered this.</p><p>“And when did we decide that I wanted it done?”</p><p>“You did not.”</p><p>“Then who did?”</p><p>The Master Mason looked puzzled by the question.</p><p>“Nobody. It became necessary.”</p><p>“When?”</p><p>“While we were doing it.”</p><p>The citizen looked at the wall.</p><p>It was, he had to admit, a very fine wall.</p><p>“And what happens now?”</p><p>“We add it to the final account.”</p><p>“Was any of this foreseeable?”</p><p>The Master Mason smiled patiently.</p><p>“My dear fellow. Anyone who has built houses for as long as I have knows that things like this happen all the time.”</p><p>The citizen was relieved.</p><p>“Excellent. Then I assume you allowed for them in the offer.”</p><p>The Master Mason stopped smiling.</p><p>“No.”</p><p>The citizen waited.</p><p>The Master Mason waited too.</p><p>“But you just said they happen all the time.”</p><p>“They do.”</p><p>“And you have been building houses for a long time?”</p><p>“Thirty-seven years.”</p><p>“Then presumably you have encountered them before.”</p><p>“Hundreds of times.”</p><p>The citizen brightened.</p><p>“So after the first few houses, you knew to expect them.”</p><p>The Master Mason frowned.</p><p>“One can never know exactly what one will find.”</p><p>“Of course not.”</p><p>The citizen pointed towards the horizon.</p><p>“A thunderstorm might wash away the foundations.”</p><p>“Indeed.”</p><p>“The ground might prove unable to bear the house.”</p><p>“Certainly.”</p><p>“The Grey Mountains might move during the night.”</p><p>The Master Mason considered this.</p><p>“They have been known to.”</p><p>“Those things I understand. We cannot put everything the world might do into an offer.”</p><p>The Master Mason nodded approvingly.</p><p>“But we are not talking about moving mountains.”</p><p>The Master Mason stopped nodding.</p><p>“We are talking about mortar.”</p><p>“Yes.”</p><p>“Scaffolding.”</p><p>“Yes.”</p><p>“And preparing stones before laying other stones on top of them.”</p><p>“That is a somewhat simplified description of masonry.”</p><p>“I am a somewhat simple customer.”</p><p>The Master Mason sighed.</p><p>“These things depend on circumstances.”</p><p>“Circumstances which occur often?”</p><p>“Very often.”</p><p>“And which an experienced mason knows about?”</p><p>“Obviously.”</p><p>“And which therefore become less surprising with every house he builds?”</p><p>The Master Mason looked increasingly uncomfortable.</p><p>“That is not how offers work.”</p><p>The citizen had suspected they might eventually arrive at this point.</p><p>“How do offers work?”</p><p>The Master Mason straightened himself.</p><p>“We offer the work that can be determined beforehand.”</p><p>“And the rest?”</p><p>“Is charged according to actual expenditure.”</p><p>“Even when experience tells you that there will almost certainly be a rest?”</p><p>“One cannot know how much.”</p><p>The citizen thought about this.</p><p>“Could you estimate it?”</p><p>“Of course.”</p><p>The answer came so quickly that both men were briefly surprised by it.</p><p>“How much?”</p><p>The Master Mason looked at the wall, then at the foundations, and finally at the scaffolding.</p><p>“For a house like this?”</p><p>“Yes.”</p><p>“Perhaps another thousand crowns. Fifteen hundred if the eastern wall is worse than it looks.”</p><p>The citizen stared at him.</p><p>“So when you offered four thousand crowns, you expected the work might actually cost five thousand?”</p><p>“Perhaps five and a half.”</p><p>“And you did not think this worth mentioning?”</p><p>“It was not part of the offer.”</p><p>There it was again.</p><p>The citizen was beginning to suspect that an offer was a very particular kind of document.</p><p>It described with considerable precision what the craftsman was willing to promise before work began, it did not necessarily describe with similar precision what the craftsman expected the work to cost.</p><p>“Could you have offered five and a half thousand crowns?” the citizen asked.</p><p>“Certainly.”</p><p>“Would that have covered what you reasonably expected?”</p><p>“Probably.”</p><p>“Then why didn’t you?”</p><p>The Master Mason looked genuinely astonished.</p><p>“Because then you would have hired the other mason.”</p><p>The citizen could find no fault with this reasoning.</p><p>Which troubled him considerably.</p><p>“So the other mason offered four thousand eight hundred.”</p><p>“Yes.”</p><p>“And the third asked for five thousand six hundred.”</p><p>“A ridiculous offer.”</p><p>“Why?”</p><p>“Far too expensive.”</p><p>“How much will my wall cost?”</p><p>The Master Mason inspected it again.</p><p>“Five thousand four hundred, perhaps five and a half.”</p><p>The citizen looked down the street, in the direction from which the third mason had departed.</p><p>“Then his offer was remarkably accurate.”</p><p>“That is hardly the point.”</p><p>“What is the point?”</p><p>“The point is that nobody would hire a mason who offered five thousand six hundred crowns when another offered the same work for four thousand.”</p><p>“But they did not offer the same work.”</p><p>The Master Mason shook his head.</p><p>“Of course they did. Otherwise you could not compare the offers.”</p><p>The citizen thought this was perhaps the most interesting thing he had learned about comparison.</p><p>Apparently, two offers could only be compared if they described the same thing.</p><p>And if one of them happened to include more of what would actually be required, this made it less competitive.</p><p>Accuracy, it seemed, had become a commercial disadvantage.</p><p>Fortunately, the Kingdom had anticipated disputes arising from such matters.</p><p>It had anticipated almost all disputes.</p><p>There were Royal Acts governing additional work, Guild Rules governing measurement, Ordinances concerning materials, provisions for unforeseen conditions, exceptions for necessary work, special exceptions for urgently necessary work, and several particularly sophisticated paragraphs concerning work that had become necessary as a consequence of other necessary work.</p><p>There were rules</p><ul><li>specifying when a craftsman had to inform a citizen.</li><li>specifying when he did not.</li><li>specifying what could be charged without prior agreement.</li><li>defining circumstances under which prior agreement could reasonably have been considered impractical.</li><li>explaining how to calculate the reasonable cost of work that nobody had explicitly ordered.</li></ul><p>The citizen spent several evenings reading them.</p><p>The more he read, the more impressed he became.</p><p>Almost every imaginable situation had been considered, almost every conceivable danger had a paragraph, almost every paragraph had an exception.</p><p>And almost every exception referred to another paragraph.</p><p>Eventually he visited the Office of Proper Construction and spoke to a clerk.</p><p>“I have a simple question.”</p><p>The clerk looked concerned.</p><p>Simple questions were notoriously difficult to process.</p><p>“My mason performed work that was not in his offer.”</p><p>“That can happen.”</p><p>“He says it was necessary.”</p><p>“That can also happen.”</p><p>“He did not ask me before doing it.”</p><p>“There are circumstances under which this is permitted.”</p><p>“He knew before beginning that work of this kind would probably be necessary.”</p><p>“That does not necessarily mean he knew precisely which work would be necessary.”</p><p>“He could estimate its likely cost.”</p><p>“An estimate is not certainty.”</p><p>“I understand.”</p><p>The clerk relaxed slightly.</p><p>“So I would like to know something else.”</p><p>“Yes?”</p><p>“What was I supposed to learn from the offer?”</p><p>The clerk blinked.</p><p>“The price.”</p><p>“But it wasn’t the price.”</p><p>“The offered price.”</p><p>“What is the difference?”</p><p>“The offered price is the price of the work contained in the offer.”</p><p>“And the work not contained in the offer?”</p><p>“That is additional work.”</p><p>“Even when it was reasonably foreseeable?”</p><p>“That depends.”</p><p>“On what?”</p><p>The clerk reached behind him and removed a volume of considerable size from a shelf.</p><p>“This.”</p><p>The citizen looked at it.</p><p>It was entitled *Royal Commentary on Foreseeable and Unforeseeable Necessities in the Execution of Proper Works, Volume III*.</p><p>There were four other volumes beside it.</p><p>The citizen did not open it.</p><p>Instead he asked: “Would it not be simpler to require the craftsman to tell me what he reasonably expects the whole thing to cost?”</p><p>The clerk looked at him with the patient kindness reserved for people unfamiliar with the complexities of professional affairs.</p><p>“But he cannot know that.”</p><p>“He seems remarkably good at knowing it once the work is finished.”</p><p>“That is because then the work has been done.”</p><p>“And remarkably good at estimating it once I have already hired him.”</p><p>“That is professional experience.”</p><p>“Could he use some of that experience before I hire him?”</p><p>The clerk hesitated.</p><p>“There are regulations concerning offers.”</p><p>“I know.”</p><p>The citizen looked at the shelves.</p><p>There were hundreds of them.</p><p>And slowly he began to understand.</p><p>The rules had been written to protect citizens from every conceivable way in which building work might depart from expectation. But somewhere in the attempt to describe every conceivable departure, they had become extraordinarily good at explaining why each departure was permissible.</p><p>There was a rule for everything.</p><p>Which meant there was almost always a rule for the thing that had just happened.</p><p>The craftsmen had not created this system, they had merely become very good at living in it. And they had learned something important about competition. A craftsman who included every reasonably foreseeable necessity in his offer appeared expensive, one who left them out appeared competitive.</p><p>Both might eventually build the same wall.</p><p>Both might eventually charge the same number of crowns.</p><p>But only one of them was likely to be hired.</p><p>Thus the craftsmen of the Releasable Kingdom had achieved something remarkable.</p><p>They had made optimism competitive.</p><p>And accuracy expensive.</p><p>A few days later, the Witch encountered the third mason at the edge of the market.</p><p>He was buying rope.</p><p>The Witch had always regarded rope as one of the more trustworthy products sold in the Releasable Kingdom. It was usually sold by length, and very little additional rope became unexpectedly necessary after one had paid for it.</p><p>“You offered five thousand six hundred crowns for a wall,” she said.</p><p>The mason looked at her.</p><p>“I did.”</p><p>“Another mason offered four thousand.”</p><p>“I know.”</p><p>“He got the work.”</p><p>“I know that too.”</p><p>“It will cost about five and a half.”</p><p>The mason considered this for a moment.</p><p>“That sounds about right.”</p><p>The Witch narrowed her eyes.</p><p>“How did you know?”</p><p>“I measured it.”</p><p>“So did the others.”</p><p>“I have no doubt.”</p><p>“And you inspected the same stones?”</p><p>“Yes.”</p><p>“The same ground?”</p><p>“Yes.”</p><p>“The same wall?”</p><p>“There was only one.”</p><p>The Witch liked him.</p><p>“Then why was your number different?”</p><p>The mason put down the rope.</p><p>“Because I have built walls before.”</p><p>This was not the answer she had expected.</p><p>“So has the man who got the work.”</p><p>“Thirty-seven years, I believe.”</p><p>“Then he should be better at it than you.”</p><p>“He probably is.”</p><p>The Witch waited.</p><p>The mason sighed.</p><p>“Building the wall and offering to build the wall are different crafts.”</p><p>This interested her.</p><p>He took a small notebook from his coat.</p><p>It was worn at the edges and considerably less impressive than *The Proper Measurement of Works and Materials*.</p><p>“I keep records.”</p><p>“Of what?”</p><p>“Of what I thought would happen. And what happened instead.”</p><p>The Witch opened the notebook.</p><p>There were houses.</p><p>Many houses.</p><p>Beside each were numbers. Stone. Mortar. Timber. Preparation. Additional supports. Hours. Things discovered after opening walls. Things expected but not needed. Things nobody had expected at all.</p><p>“You keep all this?”</p><p>“How else would I become better at my work?”</p><p>The Witch turned several pages.</p><p>“Your estimates became larger.”</p><p>“More accurate.”</p><p>“That is not generally considered an improvement in an offer.”</p><p>“No.”</p><p>“Why?”</p><p>“Because citizens compare the first number.”</p><p>“And craftsmen know this?”</p><p>“Of course.”</p><p>“So the safest way to win the work is to make the first number small.”</p><p>“One of the safest.”</p><p>“And then discover reality afterwards.”</p><p>The mason smiled.</p><p>“Reality has excellent timing.”</p><p>The Witch continued through the notebook.</p><p>Something bothered her.</p><p>“There are many additional costs.”</p><p>“Yes.”</p><p>“But very few additional savings.”</p><p>The mason nodded.</p><p>“That was one of the first things I learned.”</p><p>“Why?”</p><p>“Because risk is not symmetrical.”</p><p>The Witch looked pleased.</p><p>She enjoyed sentences that made obvious things sound suspicious.</p><p>The mason pointed towards a nearby house.</p><p>“Suppose I estimate the work required to repair that wall. What can happen?”</p><p>“It might require less work.”</p><p>“It might.”</p><p>“It might require more.”</p><p>“Very easily.”</p><p>“The stones might be better than expected.”</p><p>“Yes.”</p><p>“Or much worse.”</p><p>“Yes.”</p><p>“You might discover that one section does not need replacing.”</p><p>“A small saving.”</p><p>“Or discover that the timber behind it is rotten.”</p><p>“A large expense.”</p><p>The Witch nodded.</p><p>“So the pleasant surprises have a floor.”</p><p>“Eventually there is no work left to save.”</p><p>“But the unpleasant ones”</p><p>“Can become remarkably creative.”</p><p>The Witch smiled.</p><p>The Grey Mountains had taught her something similar.</p><p>“So what do you do?”</p><p>“If something is reasonably foreseeable, I calculate it in.”</p><p>“Even if it might not be necessary?”</p><p>“Especially then.”</p><p>“That makes your offer more expensive.”</p><p>“It makes my offer more useful.”</p><p>“To whom?”</p><p>“The person paying it.”</p><p>The Witch thought this an unusually unfashionable definition of usefulness.</p><p>“And if the work turns out not to be necessary?”</p><p>“I do not perform it.”</p><p>“And the citizen still pays?”</p><p>The mason looked offended.</p><p>“Why would he?”</p><p>“So the house becomes cheaper than your offer?”</p><p>“Sometimes.”</p><p>“Citizens must find that confusing.”</p><p>“They usually recover.”</p><p>The Witch imagined the peculiar distress of a citizen receiving a final account smaller than expected.</p><p>“And you can still comply with all the Royal Acts and Guild Ordinances?”</p><p>“I have to.”</p><p>“Why?”</p><p>“I rather enjoy being a mason.”</p><p>He put the notebook back into his coat.</p><p>“The Guild can remove my allowance to practise the craft. The rules tell me what I must document, how I must measure, what materials I may use, what I may charge and what I must disclose.”</p><p>“And they tell you how to make a good offer?”</p><p>“No.”</p><p>“What does?”</p><p>He tapped the notebook.</p><p>“This.”</p><p>The Witch was beginning to see why the third mason had seemed unfamiliar with the customs of the Kingdom.</p><p>He obeyed its rules.</p><p>He simply did not use them as a substitute for thinking.</p><p>“And what about things you genuinely cannot estimate?”</p><p>“I don’t.”</p><p>The Witch raised an eyebrow.</p><p>This was almost certainly against some tradition.</p><p>“You simply refuse to give a price?”</p><p>“I refuse to invent one.”</p><p>“What do you give the citizen instead?”</p><p>“Evidence.”</p><p>He opened the notebook again.</p><p>“If the work is complicated, likely to change, or depends on something we cannot see until we begin, I show them what happened before.”</p><p>He pointed to several entries.</p><p>“Seven similar houses. In two, the work remained simple. Three required additional supports. One had substantial damage behind the wall. One became a considerably different project once we opened it.”</p><p>“And then?”</p><p>“I explain what we know. What we don’t. What usually happens. What has happened occasionally. And what we can learn cheaply before committing to the rest.”</p><p>“That is not much of an offer.”</p><p>“It is a different offer.”</p><p>“How different?”</p><p>“Small.”</p><p>The Witch approved immediately.</p><p>“We agree on exactly what we will investigate. We open one section. Remove three stones. Examine one beam. Whatever gives us the information we need.”</p><p>“And then you offer the whole job?”</p><p>“If we know enough.”</p><p>“And if you don’t?”</p><p>“We learn something else.”</p><p>The Witch considered this.</p><p>“So instead of making the uncertain work part of a large offer — “</p><p>“I make reducing the uncertainty the work.”</p><p>“And charge for that?”</p><p>“Of course. Learning requires craftsmen too.”</p><p>“But then the citizen might spend money and still decide not to build.”</p><p>“Yes.”</p><p>“Isn’t that waste?”</p><p>The mason shook his head.</p><p>“Discovering that you should not spend ten thousand crowns after spending two hundred is one of the cheaper things that can happen during construction.”</p><p>The Witch looked towards the citizen’s house.</p><p>The scaffolding had grown again.</p><p>“And for ordinary work?”</p><p>“If it is foreseeable, calculate it in.”</p><p>“For uncertain work?”</p><p>“Show what experience tells us.”</p><p>“For complicated work?”</p><p>“Make learning about it a separate piece of work.”</p><p>“And if reality is kinder than expected?”</p><p>The mason picked up his rope.</p><p>“Then the citizen pays less.”</p><p>The Witch smiled.</p><p>“An interesting business model.”</p><p>“It has disadvantages.”</p><p>“You lose offers.”</p><p>“Frequently.”</p><p>“Because you appear expensive.”</p><p>“Yes.”</p><p>“Even though you may not be.”</p><p>“Yes.”</p><p>“And the craftsmen whose offers appear cheaper eventually charge roughly what you offered?”</p><p>“Quite often.”</p><p>The Witch watched him tie the rope to his cart.</p><p>“So accuracy makes you less competitive.”</p><p>“Only until someone has built enough houses.”</p><p>That evening, the Witch returned to the citizen.</p><p>She found him studying his original offer beside the latest account.</p><p>The two documents had very little in common except the name of the mason.</p><p>“I have found your expensive mason,” she said.</p><p>The citizen looked up.</p><p>“The one I couldn’t afford?”</p><p>“The very one.”</p><p>“What did he say?”</p><p>The Witch placed a small piece of parchment on the table.</p><p>On it she had written three sentences.</p><blockquote><strong>If it is foreseeable, calculate it in.</strong></blockquote><blockquote>If it is uncertain, show what experience has taught you.</blockquote><blockquote>If uncertainty itself is the problem, sell learning before selling the solution.</blockquote><p>The citizen read them twice.</p><p>“Is that allowed?”</p><p>“Oh yes,” said the Witch.</p><p>“That may be the strangest part.”</p><p><a href="https://the-effective-software-engineer.com/chronicles/#__tabbed_1_2">Read more.</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e22221001220" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Obvious Saturday: AI Does Not Think Like You Do]]></title>
            <link>https://medium.com/@steell0815/obvious-saturday-ai-does-not-think-like-you-do-bb630c64335b?source=rss-d51fc814c089------2</link>
            <guid isPermaLink="false">https://medium.com/p/bb630c64335b</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[obvious]]></category>
            <dc:creator><![CDATA[Stefan Ellersdorfer]]></dc:creator>
            <pubDate>Sat, 19 Sep 2026 06:01:01 GMT</pubDate>
            <atom:updated>2026-09-19T06:20:13.520Z</atom:updated>
            <content:encoded><![CDATA[<p>AI</p><ul><li>is hallucinating.</li><li>is getting smarter.</li><li>understands our codebase.</li><li>decided to refactor the module.</li><li>was lazy and skipped the tests.</li><li>lied.</li><li>replaced three developers.</li></ul><p>We have become remarkably comfortable talking about artificial intelligence as if there were a little person living somewhere behind the API, although there isn’t.</p><p>And this distinction matters much more than it might appear.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DMPts3NNXDha1_Jeo6Ondg.jpeg" /><figcaption>obvious saturday — a collection of observations</figcaption></figure><h3>Useful Metaphors Become Dangerous Models</h3><p>The human language is full of metaphors. We tend to say that a computer <em>remembers</em>. A service <em>listens</em>. A process <em>dies</em>. A program <em>calls</em> another program.</p><p>Nobody seriously believes that a process experiences death.</p><p>With AI, however, the boundary has become surprisingly blurry.</p><p>We say that an AI <em>thinks</em>, <em>understands</em>, <em>reasons</em>, <em>hallucinates</em>, <em>knows</em>, <em>forgets</em>, <em>lies</em>, becomes <em>confused</em> or is simply <em>lazy</em>.</p><p>These words are convenient descriptions of observable behavior, but they are poor descriptions of the mechanism producing it.</p><p>And that becomes important when the people using these words are making decisions about how software should be engineered, how organizations should be structured, and which responsibilities can safely be delegated to machines.</p><p>A metaphor is useful until we start designing organizations around it.</p><h3>No hallucinations</h3><p>Take perhaps the most established example: <em>hallucination</em>.</p><p>It is a wonderfully memorable term. and also deeply anthropomorphic.</p><p>A human hallucination is a perceptual experience occurring without the corresponding external stimulus. It happens to something capable of perception and subjective experience.</p><p>A language model does not need either to produce an incorrect statement.</p><p>It generates an output from a computational process involving statistical relationships, supplied context, system instructions, retrieved information, tools and other machinery surrounding the model.</p><p>Sometimes that process produces something that corresponds remarkably well with reality, sometimes it doesn’t.</p><p>Calling the latter a hallucination gives the failure a strangely human character, but the machine didn’t see something that wasn’t there. It rather produced an output for which our engineering system lacked sufficient mechanisms to establish whether it was true.</p><p>That is a very different problem.</p><p>And importantly, it suggests very different solutions.</p><h3>Probability != Stupidity</h3><p>The opposite simplification isn’t particularly useful either.</p><p>AI is not <em>just autocomplete</em>.</p><p>That phrase may have been useful for explaining an important aspect of language models, but it now hides almost as much as anthropomorphism does. Modern AI systems combine extraordinarily large learned representations with context, retrieval, tools, computation and increasingly sophisticated mechanisms for producing and evaluating intermediate results.</p><p>Their behaviour can therefore be astonishingly capable, but this capability does not imply consciousness. Nor does an impressive answer imply understanding in the human sense.</p><p>This distinction is essential precisely because these systems are so impressive.</p><p>If a calculator occasionally produced plausible but incorrect answers, nobody would say:</p><p><em>Well, it seemed very confident.</em></p><p>We would consider the calculator broken.</p><p>With generative AI, plausible output is part of the mechanism and that changes how we must engineer systems around it.</p><h3>The Dangerous Part Is Fluency</h3><p>Large language models were trained extensively on language - the medium humans use to expose thought.</p><p>That makes the result overwhelming persuasive.</p><p>The system doesn’t merely produce an answer, it can explain, rephrase and defend it. Challenge your objection. Produce examples.</p><p>Write a test and apologize for being wrong.</p><p>Confidently explain why its newly generated answer is now correct.</p><p>This creates something technologically remarkable and psychologically treacherous: <strong>The interface looks remarkably similar to understanding.</strong></p><p>For casual use, that distinction may not matter very much.</p><p>For software engineering, it matters enormously.</p><h3>AI Changes Software Engineering</h3><p>This is why I find the question</p><blockquote>Will AI replace software developers?</blockquote><p>less interesting than it initially appears.</p><p>If by software development we mean translating reasonably well-defined solutions into code, then AI can already automate a remarkable amount of that work, and it almost certainly will become increasingly more efficient at it.</p><p>But software engineering was never primarily valuable because humans were particularly good at typing syntax into files.</p><p>Software exists because we expect it to change.</p><ul><li>Businesses change.</li><li>Users teach us things.</li><li>Regulations change.</li><li>Dependencies change.</li><li>Platforms disappear.</li><li>Security vulnerabilities emerge.</li><li>Architectures encounter conditions nobody anticipated.</li><li>Occasionally we discover that the beautifully implemented solution solves the wrong problem.</li></ul><p>AI accelerates our ability to make changes inside this environment and that makes engineering discipline more important, not less.</p><h3>Faster Requires Better Evidence</h3><p>Suppose an AI agent can implement a feature in three hours instead of a developer taking three days.</p><p>Wonderful.</p><p>Now comes the interesting question:</p><p><em>How do we know that the result is correct?</em></p><p>Not whether…</p><p>… the code looks reasonable.</p><p>… another language model likes it.</p><p>… the agent tells us that all requirements have been satisfied.</p><p>What evidence does the engineering system provide?<br>Does the behavior satisfy executable acceptance criteria?<br>Do the tests establish the relevant properties?<br>Does the software compile, does static analysis pass?<br>Are dependencies acceptable?<br>Are security constraints satisfied?<br>Can the artifact be deployed?<br>Does it actually run?<br>Can we observe its behaviour in production?<br>Can we safely change it again tomorrow?</p><p>Those questions existed before generative AI.</p><p>AI simply makes it possible to produce changes fast enough that weaknesses in answering them become painfully obvious.</p><h3>Build a trustworthy Engineering System.</h3><p>There is an interesting consequence here.</p><p>We should not build engineering systems around the assumption that AI will eventually become sufficiently trustworthy, nor build them around the assumption that humans are sufficiently trustworthy either.</p><p>Humans misunderstand requirements, introduce defects, forget tests, overlook security problems, confidently implement the wrong solution.</p><p>The great achievement of professional software engineering has never been eliminating human fallibility, but building systems in which fallibility can be detected quickly and cheaply:</p><ul><li>Version control.</li><li>Automated tests.</li><li>Continuous Integration.</li><li>Executable specifications.</li><li>Static analysis.</li><li>Security scanning.</li><li>Deployment automation.</li><li>Observability.</li><li>Fast feedback.</li><li>Small changes.</li><li>Releasable software.</li></ul><p>AI fits remarkably well into such an environment.</p><p>Because it does not have to think like an engineer. It can propose changes at enormous speed while a deterministic engineering system continuously establishes evidence about those changes.</p><h3>Respect the Machine for What It Is</h3><p>None of this diminishes AI - quite the opposite.</p><p>Calling AI a stochastic parrot, autocomplete, a junior developer, an artificial colleague or an emerging digital consciousness all attempt to make something unfamiliar understandable through something familiar.</p><p><strong>But we should resist that temptation.</strong></p><p>What we have created is strange enough to deserve being understood on its own terms. A computational system trained on enormous amounts of human-produced information can manipulate language, representations and tools sufficiently well to perform tasks that until very recently appeared to require human intelligence.</p><p>That is extraordinary.</p><p>We don’t need to add consciousness to make it impressive, we also don’t need to dismiss it as autocomplete to make it safe.</p><p>We need to understand its properties, exploit its strengths and engineer around its weaknesses.</p><p>Especially when we put it in charge of changing software.</p><p>Because the important question isn’t whether the AI <em>understands</em> what it has done, the important question is:</p><p><strong>Can our engineering system prove enough about what it has done that understanding is not our only line of defense?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bb630c64335b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>