<?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 Olivier Meyer on Medium]]></title>
        <description><![CDATA[Stories by Olivier Meyer on Medium]]></description>
        <link>https://medium.com/@oliviersmeyer?source=rss-f606c8886a5c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KjusjZcNG48ifMYnyHh4UA.jpeg</url>
            <title>Stories by Olivier Meyer on Medium</title>
            <link>https://medium.com/@oliviersmeyer?source=rss-f606c8886a5c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 07:09:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@oliviersmeyer/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[We’ve Seen This Factory Before]]></title>
            <link>https://medium.com/@oliviersmeyer/weve-seen-this-factory-before-3abbdd0a6c81?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/3abbdd0a6c81</guid>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 15:41:26 GMT</pubDate>
            <atom:updated>2026-10-14T15:41:26.192Z</atom:updated>
            <content:encoded><![CDATA[<p><em>What the shift from waterfall to Agile can teach us about AI-driven development.</em></p><p>I’ve been developing software long enough to have lived through several generations of development methodologies. I started my career when waterfall dominated, experienced the transition to Agile, watched Scrum become institutionalized, and now build software with AI coding agents as the industry starts talking about autonomous software factories.</p><figure><img alt="" src="https://cdn-images-1.medium.com/proxy/1*yS9zYjI8mlju-powdSbVIA.png" /></figure><p>Lately, I’ve been noticing something familiar: we’re talking about specifications again. Not just requirements or user stories, but increasingly detailed descriptions of what software should do, how it should behave, and how we’ll verify that it works correctly.</p><p>What’s interesting is that the reasons for writing detailed specifications have changed. In the waterfall era, we invested heavily in specifications partly because implementing software was expensive. Today, we’re returning to detailed specifications partly because implementation is so fast that an incomplete or incorrect specification can quickly become a substantial amount of working software.</p><p>The economics have changed, but the underlying challenge hasn’t.</p><p>We may soon be able to produce software faster than we can understand what we should produce.</p><p>Others are raising similar concerns. But I think it’s worth looking at what happened the last time we tried to organize software development around getting the specification right.</p><h3>What Agile Was Supposed to Fix</h3><p>Early in my career, we spent tremendous effort trying to get requirements right before development began. We wrote detailed requirements documents, developed use cases, created UML diagrams using tools like Rational Rose, and modeled systems from multiple perspectives. The idea was to understand as much as possible about what we were building before writing the code.</p><p>There were good reasons for this. Writing software was expensive, and changing it after significant development had occurred could be even more expensive. The more we could anticipate upfront, the less rework we’d encounter later. At least that was the theory.</p><p>The problem was that requirements were rarely as complete as we thought. Customers didn’t always know exactly what they needed. We didn’t always ask the right questions. And assumptions that seemed reasonable on paper sometimes fell apart when people saw the software working.</p><p>Worse, development could take so long that business priorities and customer needs had changed by the time we delivered. We could execute perfectly against the original specification and still miss the mark.</p><p>As Barry Boehm famously distinguished, there’s a difference between building the product right and building the right product.</p><p>Agile challenged the assumption that we needed to understand everything upfront. Instead, we could build something smaller, get feedback, learn, adjust, and repeat.</p><p>But one of Agile’s most important contributions often gets overlooked. It wasn’t just about shorter development cycles. It was about changing how people worked together.</p><p>The <a href="https://agilemanifesto.org/principles">Agile Manifesto’s principles</a> explicitly call for business people and developers to work together daily. Cross-functional teams brought product, engineering, testing, and design into a collaborative process. Engineers could challenge assumptions, ask questions, suggest alternatives, and help shape the solution. Requirements became the beginning of a conversation rather than the final word.</p><p>Of course, Agile didn’t always work that way in practice. Over time, I watched daily standups become status reports and sprint reviews turn into demonstrations of completed work, often with limited participation from business stakeholders. We began questioning whether these ceremonies were worth the time.</p><p>Sometimes that criticism was justified. But the ceremonies weren’t the point. The conversations and learning they were supposed to create were. In trying to optimize Agile for delivery, I think we sometimes lost sight of the collaboration that helped us determine what to deliver.</p><p>Many teams drifted into what John Cutler called <a href="https://medium.com/@johnpcutler/12-signs-youre-working-in-a-feature-factory-44a5b938d6a2">feature factories</a>: very good at shipping features, less focused on whether those features delivered meaningful outcomes. Across the industry, Agile teams became delivery teams, only to rediscover that we needed product teams.</p><p>That history is ironic, given how enthusiastically we’re now talking about autonomous software factories.</p><h3>New York in the Wrong Hemisphere</h3><p>Over the past year, I’ve spent a fair amount of time building software with AI coding tools, particularly Claude Code. One of those projects is GeoSheet, a spatial spreadsheet I’ve been building largely on my own, with Claude Code as my primary development tool.</p><p>The experience has been remarkable. Tasks that previously required significant engineering effort can sometimes be completed in a fraction of the time. But I’ve also noticed how much the result’s quality depends on how well I communicate what I want.</p><p>When I’ve worked with experienced engineers, they’ve often had context beyond the requirements themselves. They knew the customers, understood why we made previous decisions, and sometimes challenged what we asked them to build. AI agents can do some of that too, but I’ve found they’re often willing to make a reasonable assumption and keep going rather than stop and ask a question.</p><p>Sometimes those assumptions are correct. Sometimes they’re not.</p><p>It’s not surprising that we’re seeing renewed interest in specification-driven development. Frameworks such as <a href="https://github.blog/ai-and-ml/generative-ai/spec-driven-development-with-ai-get-started-with-a-new-open-source-toolkit/">GitHub’s Spec Kit</a> emphasize structured specifications, implementation plans, and verification. GitHub describes an iterative process, and I’ve found that clearer requirements, constraints, and acceptance criteria make a real difference.</p><p>Having spent part of my career creating detailed requirements and UML models, I can’t help noticing the similarities. But simply adding more detail doesn’t necessarily solve the underlying problem. We tried that before.</p><p>Here’s an example from GeoSheet.</p><p>In March, I added support for Well-Known Text, or WKT, a standard format for representing geometric shapes. The complication was that the geographic data I was importing used longitude-first coordinates, while GeoSheet had traditionally used latitude-first coordinates.</p><p>The specification introduced what seemed like a reasonable heuristic to distinguish between the two. If a coordinate fell outside the range of −90 to 90, it had to represent longitude. If both coordinates could be either latitude or longitude, the system assumed GeoSheet’s existing order.</p><p>The implementation shipped with 45 new tests. Those tests included New York, Los Angeles, San Francisco, Paris, and London. But the ambiguous coordinates were tested in latitude-first order. One test explicitly confirmed that ambiguous input would default to GeoSheet’s existing convention.</p><p>That distinction mattered.</p><p>Consider New York. Its longitude is approximately −74, and its latitude is approximately 41. Both values fall within the valid latitude range. A standard longitude-first WKT polygon could therefore be interpreted in the wrong order, even though tests using New York coordinates were passing.</p><p>On April 30, about five weeks after the implementation, the problem surfaced during manual testing of a new map renderer. A New York State polygon had its coordinates swapped, placing it around East Antarctica. The testing notes described the map view being pulled into the South Atlantic while trying to frame that polygon and a correctly positioned New York point.</p><p>The eventual fix included a memorable commit message: “NY state rendered in the South Atlantic.”</p><p>The tests weren’t simply missing an example. They were confirming the wrong behavior.</p><p>I had let GeoSheet’s existing coordinate convention overrule the input format’s convention.</p><p>That’s what makes this interesting to me. I had agreement between the specification, implementation, and tests, but that agreement didn’t make the assumption correct.</p><h3>When the Tests Agree With the Mistake</h3><p>Imagine an autonomous software factory where multiple specialized agents collaborate to produce software. A requirements agent develops a specification. A planning agent determines how to implement it. Development agents write the code. Testing agents generate and execute tests. Review agents examine the results.</p><p>Each agent successfully performs its assigned task. The implementation passes its tests, the reviews are completed, and everything looks good. But what happens when the original understanding of the problem was wrong? Or when the agents share the same incorrect assumption?</p><p>In traditional waterfall development, work moved through sequential handoffs. Agile attempted to address the resulting misunderstandings through cross-functional collaboration and continuous feedback. Now we’re building systems that could automate much of that work.</p><p>If one of Agile’s most important contributions was creating opportunities for people to question requirements, how do we preserve that in an autonomous software factory?</p><p>Some software-factory approaches already address this concern. <a href="https://factory.strongdm.ai/principles">StrongDM’s principles</a>, for example, emphasize end-to-end validation, holdout scenarios, and feedback from real-world behavior. They include incident replays, customer interviews, and surveys among the inputs that can help improve the system.</p><p>I think that’s encouraging. Those mechanisms can help us test more than whether the implementation agrees with the specification.</p><p>But they also bring us back to a familiar question: how do we know that our scenarios and success criteria reflect what customers actually need? And what causes us to reconsider them?</p><p>An independently maintained scenario can catch something the implementation’s own tests missed. But if the scenario and the specification share the same incorrect assumption, we still need something outside that agreement to reveal the problem.</p><p>There’s a reasonable counterargument here. If AI makes software development dramatically cheaper, does it really matter if we build something incorrectly? Can’t we just change the specification and regenerate it?</p><p>Sometimes, absolutely. One of the most exciting aspects of AI-assisted development is how quickly we can experiment, throw something away, and try again. But we have to recognize that what we’ve built is wrong before we can correct it.</p><p>If the specification, implementation, and tests share the same incorrect assumption, the system may have no reason to flag the problem. And while regenerating code may be inexpensive, discovering a mistake isn’t necessarily inexpensive. Neither is the opportunity cost of pursuing the wrong solution, exposing customers to incorrect behavior, or introducing complexity that subsequent changes must account for.</p><p>Making code cheaper to produce doesn’t necessarily make mistakes cheaper to discover.</p><p>AI coding agents can take on more of the implementation work. But we still have to understand the problem, question our assumptions, and figure out what to build.</p><h3>Designing for Questions</h3><p>I’ve seen that distinction play out in my own development process.</p><p>When I started building GeoSheet, I moved quickly, often going directly from an idea to implementation. Over time, I added more structure: written specifications and open questions to resolve before development. Pull requests also received automated reviews from Devin (by Cognition), separate from the Claude Code sessions used for development. More recently, I added evaluations that tested the AI assistant’s behavior from the user’s perspective.</p><p>In some ways, I found myself returning to practices that reminded me of earlier stages of my career. But the goal wasn’t to specify everything perfectly upfront. The goal was to create more opportunities to catch incorrect assumptions before they became embedded in working software.</p><p>I’ve also seen this process surface problems before implementation. While planning background operations for GeoSheet’s AI assistant, the specification raised sorting and undo as an open question. The follow-up analysis found that a partial undo could leave edited cells associated with the wrong records. Sorting, along with row inserts and deletes, was left out of that phase.</p><p>That’s the kind of conversation I want to make sure I’m involved in.</p><p>The point isn’t whether agents can exercise judgment. It’s whether our processes create the moments where that judgment gets used.</p><p>A process that rewards task completion will tend to produce completed tasks. We should also make room for discovering that a task needs to change, or that we shouldn’t be doing it at all.</p><h3>Building Factories That Learn</h3><p>I don’t think software factories are a bad idea. One of the most exciting possibilities is that AI could help us realize some of Agile’s original goals more fully than we ever could before.</p><p>The same speed that lets agents produce software faster could also let us put working software in front of customers sooner, test assumptions, gather feedback, and make adjustments. But only if we use that speed to accelerate learning rather than simply accelerate production.</p><p><a href="https://dora.dev/capabilities/user-centric-focus/">DORA’s guidance on user-centric focus</a> makes a similar point, explicitly warning of a feature-factory trap when AI accelerates output without sufficient attention to user needs.</p><p>If we’re not careful, we could create the most efficient feature factories we’ve ever seen.</p><p>The technology might be fundamentally different, but the organizational problem would be remarkably familiar.</p><p>I think three principles are worth carrying forward as we design these systems.</p><p><strong>Challenge assumptions using independent evidence.</strong> If every agent relies on the same incomplete specification, adding more agents doesn’t necessarily improve the outcome. We need customer evidence, established domain conventions, production behavior, or independent evaluations that can challenge the specification itself.</p><p><strong>Don’t just measure how much software we’re producing.</strong> How quickly are we validating assumptions? Are faster development cycles translating into better customer outcomes? Those questions matter as much as the number of tasks completed or pull requests merged.</p><p><strong>Someone still needs to own whether we’re building the right thing.</strong> Agents can help investigate customer needs, evaluate alternatives, and recommend priorities. They can’t be accountable for whether we’re solving a problem worth solving or whether the result creates meaningful value for customers.</p><p>That doesn’t necessarily mean adding another review agent or approval step. It means designing feedback and learning into the process rather than assuming successful execution equals success.</p><p>The problem with waterfall wasn’t that we wrote detailed requirements; it was that we put too much confidence in our ability to get them right upfront, and we often discovered too late that we hadn’t. Agile helped us recognize that building software is also a learning process.</p><p>AI gives us an opportunity to dramatically accelerate that process. We can use that speed to understand problems, challenge assumptions, validate solutions, and deliver better products.</p><p>That is the type of software factory I’d want to build.</p><p>Because the real risk isn’t that software factories will fail to produce working software.</p><p>It’s that they’ll become extraordinarily good at producing exactly what we specified, even when what we specified was wrong.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3abbdd0a6c81" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Which Apocalypse Should I Be Worried About?]]></title>
            <link>https://medium.com/@oliviersmeyer/which-apocalypse-should-i-be-worried-about-9f06322ffd85?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9f06322ffd85</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[saas]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Mon, 14 Sep 2026 02:47:39 GMT</pubDate>
            <atom:updated>2026-09-14T03:01:13.737Z</atom:updated>
            <content:encoded><![CDATA[<p><em>The fear of falling behind is pushing companies to give AI more responsibility. That may be exactly why we need to be more careful about what we let it do.</em></p><p>About eight years ago, we were at Snowshoe Mountain when my daughter Leia (yes, she’s named after <em>Star Wars</em>), who was eight at the time, made an observation:</p><p>“You know what’s good about this place? For the zombie apocalypse, zombies don’t want to climb mountains.”</p><p>I asked, “Do you worry about the zombie apocalypse a lot?”</p><p>“Not that much.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mxIZX5KDQs4mXTmZcMasqQ.png" /><figcaption>Zombies don’t want to climb mountains.</figcaption></figure><p>As best I can remember, that was probably the last time I spent much time thinking about apocalypses.</p><p>Lately, apparently, I have more than one to worry about.</p><p>For the past year or so, I’ve been living through what some have called the “SaaSpocalypse,” the idea that AI will fundamentally disrupt the software industry.</p><p>The predictions vary. AI agents will replace some of the software applications we use today. Companies will need fewer software licenses because AI will do work that previously required people. Software will become dramatically cheaper to build, allowing new competitors to appear almost overnight. Companies that don’t adapt quickly enough will be left behind.</p><p>And people are taking this seriously. Companies have changed strategies. Investors have repriced businesses. Product roadmaps have been rewritten. “AI-native” (essentially, built around AI rather than simply adding it to existing products) has become practically mandatory vocabulary. People have lost their jobs because companies decided they weren’t moving toward AI quickly enough or, perhaps just as importantly, because they weren’t giving the appearance of moving quickly enough.</p><p>So I’ve spent a fair amount of time thinking about the possibility that AI might destroy the software industry I’ve spent most of my career in. I’ve also experienced firsthand how predictions about what AI might do tomorrow can have very real consequences today.</p><p>Then I started reading warnings from people working on the most advanced AI systems. Some were warning that the race among the companies building them could push the technology forward faster than our ability to understand or control it, potentially creating an existential risk to humanity. Around the same time, my wife heard a discussion about it on the radio and started asking me questions.</p><p>Apparently, I wasn’t thinking big enough about apocalypses.</p><h3>Wait. I Thought You Said It Doesn’t Think.</h3><p>Her question was pretty reasonable.</p><p>I’ve spent a lot of time telling her that today’s generative AI doesn’t really “think,” at least not in the way we normally use that word. That’s admittedly a strange thing to say about something you can have a conversation with.</p><p>Ask a good AI model to analyze a problem, write software, explain quantum physics, create a business plan, or critique something you’ve written, and it certainly looks like it’s thinking. For a growing number of tasks, the distinction can feel almost academic. It’s close enough to thinking to be extraordinarily useful.</p><p>But that doesn’t necessarily mean it understands the world the way we do. Yann LeCun has argued for years that today’s generative AI is still missing important pieces of the kind of understanding humans develop about the world. Welch Labs did a <a href="https://youtu.be/kYkIdXwW2AE?si=leRx1Rc4S_ILoM2z">great interview with him</a> recently about his work on “world models” and why he believes they may be necessary to get there.</p><p>Which leads back to my wife’s question:</p><p><strong>If these things don’t really understand what they’re doing, how exactly are they going to kill all of us?</strong></p><p>Unfortunately, they don’t necessarily have to understand what they’re doing to be very good at doing it.</p><h3>From Adviser to Actor</h3><p>Until fairly recently, most people’s experience with generative AI was pretty simple. You asked it something. It answered.</p><p>If the answer was wrong, that could certainly be a problem. But you still had to decide what to do with it.</p><p>That’s changing.</p><p>The easiest way I know to explain an AI agent is that instead of giving an AI a question, you give it a goal and a set of allowed actions. It figures out the next step, takes it, looks at what happened, and figures out what to do next. Real systems can be considerably more sophisticated, but the important change is simple: we’ve taken something that used to give us answers and started giving it permission to act.</p><p>There’s a pretty important difference between an AI saying, “Here’s what I think you should do,” and an AI saying, “I did it.”</p><h3>A Job Application Gone Wrong</h3><p>A friend of mine recently built an agent to automate his job search. It would find jobs that looked like a match and apply to them on his behalf.</p><p>One day, he applied for a job himself. Unbeknownst to him, his agent had applied for the same job.</p><p>That was awkward enough, but there was a bigger problem. The agent hadn’t answered all of the application questions exactly the way he had. In particular, it gave a different answer about his salary requirements.</p><p>The recruiter understandably wanted to know which answer was the real one. My friend had to fess up and explain what had happened.</p><p>He didn’t get the job.</p><p>It’s a funny story now, although probably less funny to my friend at the time. But what’s interesting is that the agent hadn’t gone rogue. It hadn’t become conscious or decided to sabotage his career.</p><p>It was doing exactly the kind of thing it had been designed to do. It had a goal, instructions, access to a website, and permission to take action on his behalf.</p><p>So it did.</p><p>The problem was that the action wasn’t entirely consistent with what he wanted, and he didn’t even know it had happened.</p><p>In this case, the consequence was the loss of a job opportunity. But it’s easy to imagine the same thing happening in systems businesses are beginning to deploy today. A customer service agent issues a refund that shouldn’t have been issued. A sales agent quotes the wrong price. A coding agent changes something without anyone realizing it.</p><p>Most of those mistakes are recoverable. Some could be expensive or embarrassing. But the mechanism isn’t fundamentally different from what happened to my friend. We gave the system a goal, access to something, and permission to act. It did something we didn’t quite intend.</p><p>Now keep increasing the stakes.</p><p>What happens when the thing we’ve connected it to is a municipal water system? Air traffic? The electrical grid? A laboratory? A weapons system?</p><p>At some point, “the AI did something we didn’t expect” stops being a funny story.</p><h3>It Doesn’t Have to Hate Us</h3><p>This is where I think science fiction may have trained us to think about the problem the wrong way.</p><p>The familiar version is that the computer becomes intelligent, becomes self-aware, decides humans are the problem, and bad things happen. We’ve made a few movies about this. In <em>2001: A Space Odyssey</em>, HAL turns against the astronauts. In <em>WarGames</em>, a computer comes frighteningly close to starting a nuclear war.</p><p>But there’s another possibility that requires considerably less science fiction.</p><p>The AI doesn’t have to become evil or angry. It can simply become capable enough to take consequential actions, make a mistake, and have enough authority for that mistake to matter.</p><p>You don’t necessarily need a superintelligence plotting humanity’s demise. You need a sufficiently capable system, access to something sufficiently dangerous, an objective that produces an outcome its creators didn’t anticipate, and inadequate checks between deciding and doing.</p><p>Or you need a bad actor who deliberately doesn’t care about those checks.</p><p>That’s a much less cinematic path to disaster. It’s also one I find easier to imagine.</p><h3>When the Two Apocalypses Meet</h3><p>And this is where my two AI apocalypses start running into each other.</p><p>The fear of the SaaSpocalypse isn’t merely pushing software companies to use AI. It’s pushing them to let AI do more.</p><p>Don’t just help the customer write the email; send it. Don’t just identify the problem; fix it. Don’t just recommend what should happen next; do it.</p><p>That’s an incredibly compelling promise: faster work, fewer manual steps, lower costs, and software that doesn’t just help people do their jobs but actually does portions of the job for them.</p><p>I find all of this exciting. I use AI every day. I’ve spent much of the last year building software with it and experimenting with what these systems can do. Hell, I’m using AI models to help me write these articles. I’m hardly arguing that we should stop using them.</p><p>But every step from adviser toward actor changes the consequences of being wrong. There’s a big difference between an AI suggesting a sentence I don’t like and an AI taking an action I don’t know about.</p><p>And here’s where the two apocalypses really do collide. The fear of the first one creates pressure to do exactly the things that could increase the risk of the second: move faster, automate more, remove friction, and give AI more responsibility.</p><p>At the same time, boards want an AI strategy. Investors want an AI story. CEOs want AI roadmaps. Nobody wants to be the company that wakes up two years from now and discovers everyone else understood the transition first.</p><p>That pressure isn’t irrational. AI really is changing software. But there’s a difference between moving quickly because the technology demands it and feeling pressure to demonstrate that you’re moving quickly.</p><p>If the prevailing message is that moving too slowly poses an existential threat to your business, “Let’s make sure we understand the risks before we automate this” can start to sound suspiciously like resistance to change.</p><p>The instincts required to manage the risk can point in the opposite direction: understand how the system might fail, limit what it’s allowed to do, keep appropriate checks in place, and ensure a human is still involved when the consequences warrant it.</p><p>Companies really do need to move quickly. But we’re making the transition from adviser to actor at exactly the same time the industry is being told that moving too slowly may be an existential threat.</p><h3>Maybe WarGames Had It Right</h3><p>I’ve always thought of WarGames as a very 1980s movie: nuclear war, Cold War paranoia, giant computers with blinking lights, and a synthesized voice.</p><p>But its central idea has aged pretty well.</p><p>The dangerous decision wasn’t really creating a computer capable of playing nuclear war games. It was connecting that computer to the system capable of fighting one.</p><p>We spend enormous amounts of time debating how intelligent AI will become. Will it someday be smarter than humans? Does it reason? Does it understand? Could it become conscious?</p><p>I don’t know the answers to those questions. But there’s another one we don’t need to wait to answer:</p><p><strong>What are we willing to connect it to?</strong></p><p>We don’t need to settle the philosophical question of whether an AI truly thinks before we give it the ability to act.</p><p>We’re already doing that.</p><h3>So Which Apocalypse Should I Worry About?</h3><p>I still think AI is fundamentally reshaping the software industry. I don’t know whether that means the SaaSpocalypse actually happens. Software companies have survived supposedly existential technology transitions before, usually by changing along with them.</p><p>I also don’t know whether AI represents an existential threat to humanity. People who understand the most advanced AI systems far better than I do appear to disagree about that risk.</p><p>But I do think there’s something strange about the moment we’re in. We’re simultaneously being warned that AI is advancing so quickly that companies need to move faster or risk becoming irrelevant, and that we need to be extraordinarily careful about what we allow it to do.</p><p>Both might be true.</p><p>I don’t think the answer is to stop. I certainly intend to keep building with AI and exploring what these systems can do.</p><p>But capability isn’t the same thing as judgment. As we give AI more autonomy and connect it to systems where mistakes have greater consequences, the checks and balances probably need to get stronger rather than disappear in the name of efficiency.</p><p>Twelve months ago, I thought the AI apocalypse I needed to understand was the one coming for SaaS.</p><p>Now apparently I have another one to worry about.</p><p>Not to mention the zombie apocalypse.</p><p>Maybe Leia had the right approach.</p><p>Worry. But maybe not that much.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9f06322ffd85" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is AI Better at Avoiding “Not Invented Here” Than We Are?]]></title>
            <link>https://medium.com/@oliviersmeyer/is-ai-better-at-avoiding-not-invented-here-than-we-are-c3ffcd26ffd3?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3ffcd26ffd3</guid>
            <category><![CDATA[decision-making]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 14:09:04 GMT</pubDate>
            <atom:updated>2026-08-05T14:09:04.013Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MaIUgStOfni3QJxjDCJDhA.png" /></figure><p>This weekend I was driving to get gas when a software term popped into my head.</p><p>Whatever happened to “Not Invented Here?”</p><p>“Not Invented Here” (NIH) was one of those software discussions that always seemed settled. Don’t reinvent the wheel. Build where you differentiate. Stand on the shoulders of everything that came before.</p><p>And yet, almost every experienced software engineer has a story about the homegrown ORM, the custom messaging framework, the proprietary workflow engine, or the internal tool that slowly became another product to maintain.</p><p>I’ve seen NIH from several different perspectives over the course of my career.</p><p>At Logi Analytics, our Logi Info product was, in many ways, a wrapper and orchestration layer around numerous third-party components. That wasn’t a weakness. It was the strategy. We focused our engineering effort on the pieces that differentiated us rather than rebuilding every layer of the stack.</p><p>Later, while leading our Sales Engineering team, I encountered the same idea from the opposite direction. One of the recurring conversations with prospective OEM customers was whether they should embed Logi Info or build reporting and dashboarding themselves. The biggest point of friction usually wasn’t whether they could build it. Many absolutely could.</p><p>The customers we eventually won often weren’t the ones evaluating the decision for the first time. They were the ones who had already built reporting into their application.</p><p>The initial implementation wasn’t the problem. It was everything that followed.</p><p>Every enhancement request. Every browser change. Every export format. Every security issue. Every customer asking for another reporting feature.</p><p>Eventually, many concluded that reporting wasn’t where they wanted to spend their engineering talent. They had proven they could build it. What they hadn’t anticipated was owning it.</p><p>Years later, at PerformYard, I experienced almost the opposite philosophy. We had our own ORM. Like many growing software companies, we accumulated internal frameworks over time. As the company matured, we became more pragmatic, adopting products like PostHog for analytics and AG Grid instead of continuing to build and maintain everything ourselves.</p><p>Neither approach was inherently right or wrong. Different companies, products, and stages of growth lead to different decisions.</p><p>A lot of the discussion around AI has focused on whether companies will build more software instead of buying SaaS.</p><p>The next question is what happens after you’ve already decided to build.</p><p>Once you’ve decided to build, how does an AI coding agent decide what to build?</p><p>That’s a different question than build versus buy.</p><p>That’s when I thought back to GeoSheet (geosheet.app).</p><p>A few months ago I built GeoSheet with Claude Code. I made a few architectural decisions up front. I wanted PostHog for analytics and Stripe for billing. Beyond that, I spent very little time prescribing frameworks, libraries, or implementation details.</p><p>Mostly, I described the product I wanted to build.</p><p>Out of curiosity, I looked back through the project’s dependencies.</p><p>The project leans heavily on mature open-source infrastructure: Fastify, Drizzle ORM, React Query, Yjs, Mapbox, Zod, Zustand, React Hook Form, and many others. The custom code is concentrated almost entirely in the parts that actually make GeoSheet… GeoSheet.</p><p>Looking back, that’s remarkably similar to where many of our Logi Info OEM customers eventually landed. Building the parts that differentiated their products created value. Owning every layer of infrastructure often didn’t.</p><p>Curious whether anyone had actually studied “Not Invented Here,” I spent a little time reading.</p><p>One paper in particular, <em>Opening the Black Box of “Not Invented Here”: Attitudes, Decision Biases, and Behavioral Consequences</em>, describes NIH as a more general decision-making bias: a tendency to reject or undervalue ideas, technologies, or knowledge simply because they originated somewhere else.</p><p>There are other explanations for what happened in GeoSheet. AI coding agents have effectively “seen” millions of software projects. Reaching for Fastify, Drizzle, or React Query may simply reflect the most common successful patterns rather than any architectural philosophy.</p><p>The research describes “Not Invented Here” as a human decision-making bias. AI is trained on the output of millions of human engineering decisions.</p><p>Does AI learn the “Not Invented Here” bias… or is it learning to avoid it?</p><p><em>P.S. One last thought I had after I finished this article was that I wonder if Claude Code will ever reply to a request, “Have you thought of buying this instead of asking me to build it?” I’m guessing an ad-supported Claude might one day.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c3ffcd26ffd3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pulling on the Thread]]></title>
            <link>https://medium.com/@oliviersmeyer/pulling-on-the-thread-84d2101a4f3b?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/84d2101a4f3b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[curiosity]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Mon, 27 Jul 2026 13:01:03 GMT</pubDate>
            <atom:updated>2026-07-27T13:01:03.444Z</atom:updated>
            <content:encoded><![CDATA[<p>My wife Kristin has probably learned that when I say, <em>“I’m just going to look something up,”</em> there’s a decent chance she won’t see me again for a few hours.</p><p>If I’m using a new technology, I eventually find myself wanting to understand how it works. Not just how to use it, but why it behaves the way it does.</p><p>Like many people, I started with ChatGPT and Claude. Before long, I found myself wondering how these models actually worked.</p><p>That led me to transformers.</p><p>Then neural networks.</p><p>Then the realization that my linear algebra and calculus were more than a little rusty.</p><p>So down the rabbit hole I went.</p><p>It started with refreshing concepts I hadn’t thought about in decades. Matrices. Vectors. Gradients. Partial derivatives.</p><p>You look up one topic, which leads to another, then another, and before you know it, what you thought would be a five-minute refresher has turned into several hours.</p><p>Every now and then I hit a wall. Mathematical notation can be pretty daunting, especially when you haven’t looked at it in decades. Sometimes I realize I’m spending more time deciphering the notation than understanding the underlying concept. At that point I try to be honest with myself. If I understand enough of the bigger picture, I’ll keep moving and come back to the details another day.</p><p>Sometimes I come back a day or two later, and it makes a lot more sense than it did the first time. It was probably the same way when I was in college.</p><p>This isn’t really unique to AI.</p><p>Years ago, while I was on the SQL Server team at Microsoft, I found myself wanting to understand how the storage engine worked, how data and index pages were organized on disk, and what actually happened under the covers. Back in college, I had written a very rudimentary SQL engine as an independent study project. Now I had the opportunity to learn how it was really done.</p><p>Experiences like that have also given me an appreciation for people who can explain difficult subjects well.</p><p>Along the way, I’ve found two YouTube channels that have become favorites: <a href="https://www.youtube.com/@WelchLabs">Welch Labs</a> and <a href="https://www.youtube.com/@3blue1brown">3Blue1Brown</a>. Both have an incredible gift for making complex math and AI concepts approachable without oversimplifying them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9O8ENJbNOyUzPzAKtBg-tw.png" /></figure><p>I only wish they’d existed when I was in college. I might have done a little better in freshman Calculus and Physics II.</p><p>I loved Linear Algebra, though. In fact, I enjoyed it enough that I became my professor’s grader the following year.</p><p>I’m also a sucker for technical and business books, especially hard copies. It makes very little sense. I already subscribe to both O’Reilly and Manning Press, giving me access to all their books in electronic form, more books than I’d ever be able to read. Yet I still find myself buying physical copies.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*IgiLfLc-yXlcp2CYw7KIZQ.jpeg" /></figure><p>There’s just something about holding and paging through a well-made book in your hands. Every now and then I come across what I think of as a beautifully put together book. Not just a book with great content, but one where someone clearly cared about the illustrations, the layout, and the overall experience of learning from it.</p><p><a href="https://www.welchlabs.com/">Welch Labs</a> recently published <em>The Illustrated Guide to AI</em>, and it quickly found its way onto my bookshelf. It’s almost the size of a coffee table book, but still practical enough to read cover to cover. The illustrations are outstanding, and many of the chapters point back to the corresponding YouTube videos if you want to explore a topic in more depth. I’ve found I understand something much better when I can approach it from a few different angles. A book, a video, another author’s perspective. Each one tends to fill in a few more pieces of the puzzle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MhkqRZgUQGBrFaNxyCT6tg.jpeg" /></figure><p>This appreciation for beautifully put together books probably explains why another favorite on my bookshelf is <a href="https://a.co/d/0fTtaur8"><em>Pasta by Design</em></a>. I picked it up years ago at a Big Data conference, which probably tells you roughly how long I’ve owned it (I don’t think people call it “Big Data” anymore). The book catalogs around a hundred different pasta shapes, includes a “family tree” showing how they’re related, and uses mathematical formulas to describe their geometry. It’s one of those books that just makes me smile.</p><p>They say curiosity killed the cat.</p><p>Mine has mostly resulted in too many books, too many browser tabs, and more than a few evenings that lasted much longer than I planned.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=84d2101a4f3b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Thought I Was Taking a Detour from Product]]></title>
            <link>https://medium.com/@oliviersmeyer/i-thought-i-was-taking-a-detour-from-product-fa9079cc4aec?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/fa9079cc4aec</guid>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[career-paths]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[customer-experience]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Mon, 29 Jun 2026 00:36:12 GMT</pubDate>
            <atom:updated>2026-06-29T00:36:12.452Z</atom:updated>
            <content:encoded><![CDATA[<p><em>I never intended to become a solutions engineer.</em></p><p>At the time, I was working at Microsoft. While I enjoyed the work, my family wasn’t enjoying living in the Seattle area. An acquaintance mentioned a solutions engineering opportunity at Logi Analytics back in Northern Virginia. I enjoyed presenting, liked working with customers, and figured I’d do it for six months, get us back to Virginia, and then find my way back into product.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b6Ja8OoezMqGUBBCXxiHeg.jpeg" /><figcaption>The original name for Logi Analytics was LogiXML, hence the “X-man”</figcaption></figure><p>Six months turned into several years.</p><p>What I thought would be a temporary stop became one of the most rewarding chapters of my career. I became Logi Analytics’ first solutions engineer focused solely on our OEM business and eventually helped build and lead a team of six, supporting more than 30 sales representatives. Together, we worked on many of the company’s largest and most technically challenging opportunities.</p><p>The role fit me better than I expected. Most of my customers were software companies, and I’d spent my career building software. I could relate to the questions their engineers, architects, and product teams were asking because they were the same questions I’d asked throughout my own career. My job wasn’t to out-talk anyone or simply close the deal. It was to help both sides determine whether our product was actually the right fit.</p><p>At the time, I thought I was taking a detour from product.</p><p>I wasn’t.</p><p>I was standing on the front line of it.</p><p>For those unfamiliar with the role, solutions engineering is about much more than giving software demonstrations. You’re responsible for helping customers determine whether your product can actually solve their problem. You answer difficult technical questions, build prototypes, troubleshoot customer environments, and work toward what we call the <em>technical win</em>.</p><p>Just as importantly, you’re responsible for recognizing when the answer is no. Sales is understandably focused on winning the business. Professional services, customer success, and support ultimately have to deliver on the commitments made during the sales process. Product has to think beyond a single customer and build capabilities that benefit many customers.</p><p>Good solutions engineers sit at the intersection of all those perspectives. Closing the wrong deal might help this quarter’s numbers, but it rarely helps the customer — or your own company. That gives you a front-row seat to where products meet reality.</p><p>There are dozens of prospect visits I could talk about, but three have stayed with me for very different reasons.</p><p>The first was in Seattle.</p><p>Tom, the account manager on the opportunity, and I flew to Seattle to meet with a prospect evaluating one of our products, Logi Ad Hoc. They liked many aspects of the platform, but they had concerns about the user experience and a few quality issues they had already encountered. Because I’d developed a reputation for helping close technically complex opportunities, I was asked to lead the demonstration.</p><p>Everything was going well until I tried to close a dialog box.</p><p>As I moved my mouse toward the Close button, it moved. I tried again. It moved again. No matter what I did, I couldn’t click it. The button literally ran away from my cursor.</p><p>The prospect looked at me. I looked back at him. He smiled and said, “See what I mean?”</p><p>There wasn’t much I could say.</p><p>We went to lunch.</p><p>We didn’t get the deal.</p><p>I’ve thought about that meeting many times over the years. Not because of the bug. Software has bugs. But because everything else became irrelevant.</p><p>The prospect wasn’t evaluating our architecture, our roadmap, or our long-term vision. He was trying to close a dialog box.</p><p>The second visit couldn’t have been more different.</p><p>The product UI itself wasn’t the problem. We simply couldn’t get it installed and configured in the prospect’s environment.</p><p>The prospect was incredibly patient. They gave our account manager, Keia, and me a small conference room while we spent over an hour on the phone with people back at headquarters, trying one thing after another. At one point, I turned to Keia and said, “I’m about ready to throw this laptop against the wall.”</p><p>Looking back, I don’t think anyone was intentionally being difficult back at the home office. Engineering was balancing priorities. Product was thinking about the roadmap. Sales wanted to win the business. I was focused on one prospect sitting just down the hall, waiting for software that wouldn’t work.</p><p>That day, I learned something that has stayed with me ever since.</p><p>Installation wasn’t separate from the product. It was the product.</p><p>On a third evaluation, a large healthcare software prospect raised a security concern that our platform couldn’t address. Rather than walking away, I built an extension that solved the immediate problem. Later, our CTO learned what I’d built and decided the capability belonged in the product.</p><p>It was one of the clearest examples I’d seen of a customer-specific solution becoming part of the product.</p><p>What started as a workaround to get a deal across the finish line was shipped as a product feature.</p><p>Years later, when I found myself leading product organizations, I realized those experiences had shaped my thinking far more than I had appreciated at the time.</p><p>Solutions engineering taught me lessons that no roadmap review, planning session, or strategy meeting could. So did time spent in professional services and customer success. But solutions engineering is where you see it first, before the deal is signed, before implementation begins, before anyone has committed to anything. You’re watching your product succeed or fail in real customer environments, with no safety net. If you want to understand what your product actually is, rather than what you intended it to be, that’s where you start.</p><p>When I accepted that role at Logi Analytics, I thought I was taking a detour from product. I was wrong. I wasn’t stepping away from product. I was standing on the front line of it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fa9079cc4aec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Dad Was a Good Man]]></title>
            <link>https://medium.com/@oliviersmeyer/my-dad-was-a-good-man-8725e576a287?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/8725e576a287</guid>
            <category><![CDATA[lifelong-learning]]></category>
            <category><![CDATA[career-development]]></category>
            <category><![CDATA[personal-growth]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[mentorship]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Wed, 24 Jun 2026 17:21:09 GMT</pubDate>
            <atom:updated>2026-06-24T17:32:56.720Z</atom:updated>
            <content:encoded><![CDATA[<p>Father’s Day last weekend got me thinking about advice.</p><p>Over the past few weeks, two different people reached out after becoming managers and asked essentially the same question. One was my son. The other was someone from one of my former teams who had recently stepped into management.</p><p>Their questions got me thinking about the person I used to call when I needed advice myself.</p><p>My father.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*KQFdYO4cCEEHqAJZdXAcfA.jpeg" /><figcaption>One of my favorite photos of my dad and me. My sister took it without either of us realizing it.</figcaption></figure><p>My dad spent most of his career in business strategy and leadership roles. Over the years, he worked at IBM, ITT, US Videotel, and later at Lexmark. He was an avid reader with broad interests. History, business, technology, politics, economics — almost everything interested him.</p><p>He used to tell me that college wasn’t really about learning a specific subject. It was about learning how to learn, because you never stop learning.</p><p>At the time, I mostly filed that under “things parents say.”</p><p>Looking back, I realize he spent the rest of his life proving it.</p><p>One of my earliest memories of that mindset involved a puzzle competition in a magazine called <em>Games</em>.</p><p>The grand prize involved solving a word puzzle, and my father decided that instead of tackling it by hand, he would use his APL programming skills to try to solve it. APL stood for “A Programming Language” and made heavy use of mathematical notation and Greek symbols. It was not exactly a gentle introduction to programming for a kid.</p><p>At the time, computer access wasn’t something you took for granted. My father rented time on a computer and had a terminal in his office that printed its output on paper. In the evenings, we would sit together, entering words from the dictionary to help build the data set for his program.</p><p>He didn’t win the contest. I think he finished second.</p><p>But looking back, that probably wasn’t the most important outcome.</p><p>Without realizing it, he was introducing me to computers, programming, problem solving, and the idea that difficult problems could be approached in creative ways.</p><p>In his 80s, he decided to take a course in quantum physics simply because he wanted to understand it better. I guess he was always trying to figure out how things worked.</p><p>He was, however, really (and I mean really) bad at making PowerPoint slides, that is, unless you like an abusive use of yellow on purple for slides.</p><p>When I was younger and wrestling with a difficult decision, whether it involved work, family, or life in general, he was often the person I called. He would listen, process what I was saying, ask questions, challenge assumptions, and help me think through the situation.</p><p>Looking back, I don’t remember many specific answers he gave me. What I remember is that after talking with him, I usually understood the problem more clearly. More often than not, I arrived at an answer.</p><p>He just helped me get there.</p><p>When they asked for advice, I found myself reaching for the same ideas I always come back to: <strong>be authentic</strong>, <strong>treat people the way you’d want to be treated</strong>, <strong>focus on what matters</strong>, and <strong>do quality work</strong>.</p><p>As I thought about those ideas over Father’s Day weekend, I don’t remember my father explicitly teaching me any of them.</p><p>I don’t remember conversations about leadership frameworks, management techniques, product strategy, or organizational design.</p><p>What I remember is watching how he approached people, how he approached problems, and how he carried himself.</p><p>Looking back over my career, the decisions that still bother me are rarely the ones that failed. Products fail. Strategies fail. Hires don’t work out. Those things happen.</p><p>The decisions that stayed with me were the much rarer moments where I wasn’t fully true to my own principles. The moments where I compromised more than I should have, avoided a difficult conversation, or failed to live up to the standard I expected from myself. Those are the decisions that cost me sleep.</p><p>Looking back, I think that’s another thing I learned from him. Not that you’ll always be right. Not that you’ll always succeed. But the decisions you can live with are usually the ones where you stayed true to your principles.</p><p>He listened. He learned. He cared about doing things well. And he tried to do the right thing.</p><p>The older I get, the more I think the most important lessons are learned that way — not through instruction, but through example, through years of conversations that seemed ordinary at the time.</p><p>My dad was a good man.</p><p>Looking back, I suspect many of the things I value in leadership have less to do with management and more to do with the example he set.</p><p>If some of those ideas now find their way into the advice I give my son and others beginning their own leadership journeys, then perhaps that is one more way those conversations continue.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8725e576a287" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Fort Minor Taught Me About Product Leadership]]></title>
            <link>https://medium.com/@oliviersmeyer/what-fort-minor-taught-me-about-product-leadership-262dabdc46e6?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/262dabdc46e6</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[application-modernization]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Sun, 14 Jun 2026 17:10:29 GMT</pubDate>
            <atom:updated>2026-06-14T17:10:29.455Z</atom:updated>
            <content:encoded><![CDATA[<h3>10% Luck, 20% Skill, 50% Pain</h3><p>Years ago, while leading a platform modernization effort at Logi Analytics, I drew a pie chart on my whiteboard.</p><p>It wasn’t based on customer research, market analysis, or anything I learned in business school. It was based on a song I liked.</p><p>The song was <em>Remember the Name</em> by Fort Minor, and the pie chart was simply the percentages from the lyrics:</p><ul><li>10% Luck</li><li>20% Skill</li><li>15% Concentrated Power of Will</li><li>5% Pleasure</li><li>50% Pain</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lRRBEZr_0nbbLcYS7t6PHg.jpeg" /><figcaption>The original NGP whiteboard from my office (one of many).</figcaption></figure><p>At the time, it was a combination of humor and hard-earned experience. The project was already teaching me that successful transformations require far more than technology alone.</p><p>In hindsight, it may have been one of the more accurate descriptions of product leadership I’ve encountered.</p><h3>The 10-Year Rule</h3><p>The story actually started years earlier.</p><p>While at Microsoft, I led a project exploring how software companies approached platform modernization. The software industry was changing rapidly. Open source databases such as MySQL were gaining traction, web applications were changing customer expectations, and new architectures and deployment models were emerging.</p><p>As part of the project, I spoke with software companies that had gone through significant platform transitions. One pattern kept appearing.</p><p>The products weren’t failing. In many cases, they were quite successful. The challenge was that years of success had accumulated into something else: architectural constraints, customer expectations, technical debt, and assumptions that no longer aligned with market direction.</p><p>The exact details varied from company to company, but the pattern felt familiar. At some point, many teams found themselves wrestling with the same question: Can we continue evolving what we have, or is it time for something more fundamental?</p><p>The timing wasn’t identical, but it often seemed to happen around the ten-year mark. I started thinking of it as the ten-year rule — not a law, more of an observation.</p><p>Years later, I found myself confronting the same pattern again.</p><h3>What Is NGP?</h3><p>Before moving into Product, I had spent several years as a sales engineer at Logi Analytics, where I built and scaled our OEM Sales Engineering practice. At its peak, we supported nearly thirty sales representatives and many of the company’s largest and most strategic opportunities.</p><p>Along the way, I had the opportunity to meet and learn from a wide range of software companies and product teams. As they embedded analytics into their own products, I got to see how they approached architecture, extensibility, user experience, scalability, and platform evolution.</p><p>Over time, patterns started to emerge. I also knew our own product well enough to recognize where it struggled.</p><p>Architecturally, it reflected decisions from an earlier era of software. The user experience reflected assumptions that were becoming increasingly dated. We had continued to evolve successfully, but many of the workarounds required to win larger opportunities suggested we were approaching the limits of incremental improvement.</p><p>When the company launched an initiative called NGP, short for Next Generation Platform, I stepped out of Sales Engineering and into Product to lead the effort.</p><p>Almost immediately, the organization divided into camps. One group believed the existing platform could continue evolving. Another believed more fundamental change was necessary. There was also a fair amount of skepticism. For some members of the legacy team, NGP felt less like an evolution and more like something being imposed upon them.</p><p>The debates often centered on architecture, frameworks, migration strategies, and emerging technologies. Angular was still relatively new at the time, and engineers had strong opinions about what technologies should define the future platform.</p><p>Those discussions mattered, but they weren’t the hardest part.</p><p>The hardest part was a deceptively simple question that kept surfacing:</p><p><strong>What is NGP?</strong></p><p>At first, I treated it as a product definition problem. Over time, I started to realize people were asking something deeper: Why are we doing this? What problem are we really solving? What happens to everything we’ve already built? What will the future look like? Where do I fit in that future?</p><p>Around that time, I worked with someone from our marketing team to create a video about NGP. Rather than explaining the platform ourselves, we decided to walk around the company and ask people a simple question:</p><p><strong>What is NGP?</strong></p><p>The answers were all over the map.</p><p>Some people talked about technology. Others discussed customers, architecture, innovation, or the company&#39;s future. Everyone seemed to see a different aspect of the initiative.</p><p>The exercise taught me something important. People weren’t really asking for a definition of NGP. They were trying to understand what the change meant for them, their teams, and the company&#39;s future.</p><p>At the time, I remember telling my colleague that I kept thinking about <em>Remember the Name</em>. I thought it was mostly because I liked the song&#39;s energy.</p><p>Years later, I came to realize it was probably something else. The technology was only part of the conversation. NGP had become a vehicle for a much broader discussion about where the company was headed and how we would get there.</p><p>Most transformation efforts eventually converge on some version of the same questions. Today, many organizations are asking them about AI.</p><p>The technologies change.</p><p>The questions rarely do.</p><h3>The Percentages</h3><p>The whiteboard pie chart wasn’t entirely wrong.</p><h3>10% Luck</h3><p>I was fortunate.</p><p>I had spent years working with prospects and customers before moving into Product. I wasn’t trying to imagine where the platform struggled. I had lived those conversations. I had also spent years working with software companies building their own products, which gave me exposure to a broad range of approaches, challenges, and architectural decisions.</p><p>I had spent time studying platform transitions earlier in my career, and by the time NGP arrived, I already had a working theory about what happens to successful software platforms as they age.</p><p>Luck often looks like preparation in hindsight. Sometimes it simply means being in the right place to recognize a problem when it appears.</p><h3>20% Skill</h3><p>Skill matters too.</p><p>NGP required all the usual product and technical disciplines. We spent countless hours discussing architectures, frameworks, migration strategies, extensibility, and user experience. We evaluated alternatives, debated tradeoffs, and tried to make decisions that would position the platform for the next decade.</p><p>Product leadership requires understanding customers, technology, markets, and business realities. It requires making tradeoffs, communicating clearly, and navigating uncertainty while still making decisions.</p><p>The technical aspects were difficult. The product decisions were difficult. Neither turned out to be the hardest part.</p><h3>15% Concentrated Power of Will</h3><p>This may be the most underrated percentage in the song.</p><p>Large transformations require persistence. You don’t communicate the vision once; you communicate it hundreds of times. You answer the same questions repeatedly and revisit the same concerns repeatedly.</p><p>Many people think product leadership is about creating a vision. In practice, much of the work is helping others see it.</p><h3>5% Pleasure</h3><p>The song only allocates 5% to pleasure.</p><p>That may sound pessimistic until you’ve spent enough time building products.</p><p>The pleasure is real. It’s the moment when a customer understands the vision. It’s the design review where people begin to see the future taking shape. It’s the launch, the customer success story, or the deal won because of something your team built.</p><p>Those moments matter. They are part of what makes the work rewarding. They are also brief.</p><p>A launch is rarely the end of the story. More often, it marks the beginning of the next chapter. New customer requests emerge. Competitors respond. New opportunities appear. Success creates its own set of challenges.</p><p>Five percent feels about right.</p><h3>50% Pain</h3><p>This was the largest slice on my whiteboard, and it wasn’t there by accident.</p><p>What struck me over time was that many of our most difficult discussions weren’t really about technology.</p><p>We would spend an hour debating architecture and leave, realizing that the disagreement had very little to do with architecture at all. Some people worried about risk. Others worried about disrupting existing customers or abandoning years of investment. Still others were excited by the possibilities of what a modern platform could enable.</p><p>The conversation sounded technical, but the underlying concerns were often organizational, financial, or emotional.</p><p>The same questions kept resurfacing in different forms. How much should we preserve? How much should we change? How much risk were we willing to accept?</p><p>Every answer seemed to create a new question.</p><p>Over time, I came to appreciate that the hardest part of a transformation isn’t usually the technology. It’s helping an organization navigate uncertainty while continuing to run a successful business. It’s helping people let go of assumptions that have served them well for years. It’s carrying a vision that isn’t yet obvious to everyone around you.</p><p>Technology is hard. But technology is often the easy part.</p><p>Helping people align around a different future is harder.</p><h3>The Marathon</h3><p>Having run two Marine Corps Marathons, I’ve come to believe that product transformations have more in common with endurance sports than most people realize.</p><p>Everyone is enthusiastic at the starting line. The challenge is sustaining momentum mile after mile when the finish line is no longer visible.</p><p>Product transformations are much the same. The excitement comes early. The difficult part comes later — when progress is slower than expected, debates continue longer than expected, priorities shift, and the destination remains frustratingly far away.</p><p>After roughly a year and a half working on NGP, I was ready for a change.</p><p>I still believed in the vision and the direction we were pursuing, but transformation efforts are demanding. They require persistence, patience, and an enormous amount of organizational energy.</p><p>When the opportunity to join Zoomdata emerged, it felt like a chance to continue pursuing many of the same ideas from a different starting point.</p><h3>The Finish Line</h3><p>Over the years, I’ve occasionally received unexpected messages from former colleagues. Someone mentions a roadmap document I wrote years earlier. Someone references a strategy presentation. Someone tells me they’re reading a Jira ticket I wrote five or ten years ago.</p><p>The first time it happened, I was surprised. Eventually, I started to realize that product leadership often leaves artifacts behind. Sometimes they’re roadmaps, architecture decisions, or strategy documents. Sometimes they’re teams, processes, or ways of thinking that outlast the people who introduced them.</p><p>Some of the most meaningful work I’ve done continued long after I left. NGP was one example.</p><p>More recently, I saw similar patterns emerge during the early stages of PerformYard’s AI journey. Different technology. Different market. Many of the same questions: Why are we changing? What does the future look like? How do we get there?</p><p>I won’t be there for every mile of that journey either.</p><p>That’s okay.</p><p>I’ve come to believe that the measure of impact isn’t whether you’re present at the finish line. Sometimes it’s whether you helped an organization start running in the right direction.</p><p>I think that’s what the pie chart on my whiteboard was trying to tell me all those years ago. The hardest part of product leadership was never the technology. It was helping people navigate change.</p><p>And if the pie chart got anything right, it was the size of that slice.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=262dabdc46e6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Technology Is Hard, but It Usually Turns Out to Be the “Easy” Part]]></title>
            <link>https://medium.com/@oliviersmeyer/technology-is-hard-but-it-is-usually-the-easy-part-0f9419d26d10?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/0f9419d26d10</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[tech-history]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Tue, 09 Jun 2026 02:23:54 GMT</pubDate>
            <atom:updated>2026-06-09T02:34:58.758Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Why being right about a technology trend is not the same as capturing the value it creates. Reflections on Minitel, the telecom boom, cloud computing, big data, and AI.</em></p><p>In a recent <a href="https://medium.com/@oliviersmeyer/from-minitel-to-ai-living-through-seven-technology-shifts-9a1e8de536ec">article</a> about Minitel and Station F, I wrote about how successful technologies eventually become infrastructure. After publishing it, I found myself thinking about another pattern I’ve seen over the course of my career.</p><p>I’ve been fortunate to have a front-row seat to several major technology waves. From online services and the early Internet to cloud computing, big data, and now AI, I’ve spent more than three decades watching new technologies emerge, mature, and eventually become part of everyday life. Looking back, what strikes me is that the broad technology trends were often easier to identify than the specific ways they would create value and who would ultimately capture it.</p><p>People wanted online services. Businesses wanted more connectivity. Computing resources moved into the cloud. Data became a strategic asset. AI is increasingly becoming part of products, workflows, and business processes. None of those developments feel particularly surprising in hindsight. What surprises me is how often the companies, business models, and investors that seemed best positioned at the beginning of a technology wave turned out not to be the biggest beneficiaries once that wave matured.</p><p>Early in my career, I worked for US Videotel and a related French company that brought Minitel technology to the United States. For those unfamiliar with it, Minitel was a French online service platform that allowed users to search directories, exchange messages, make reservations, conduct commerce, and access information services years before the Web became mainstream.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/826/1*VofjXVQnw45gAn8vBaAf0A.png" /></figure><p>When people talk about Minitel today, it is often portrayed as a dead end that was eventually displaced by the Internet. Looking back, I think that misses part of the story. There was nothing fundamentally wrong with the vision. People clearly wanted online services, and many of the use cases we were pursuing eventually became part of everyday life. What I failed to appreciate at the time was how important the underlying architecture would become. While Minitel supported a surprisingly rich ecosystem of service providers, it remained tied to a particular network model, interface, and device. The Internet ultimately proved far more adaptable with its decentralized approach. The future we were imagining largely arrived, just not through the architecture many of us expected.</p><p>Later in my career, I joined Wam!Net during the telecom boom. It was a period when enormous amounts of capital were flowing into communications infrastructure, and companies such as Global Crossing were investing billions of dollars based on the belief that Internet traffic would continue to grow dramatically.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LsEEv-ULkIdFE9QZVbaY7g.png" /><figcaption>Wam!Net sneakers represented the move away from sneakernet.</figcaption></figure><p>Looking back, that prediction was largely correct. Internet traffic exploded, businesses digitized, and connectivity became essential. What proved much harder to predict was the economic outcome. Infrastructure was built faster than demand could absorb it, prices fell dramatically, and many investors failed to realize the returns they expected. I don’t think the industry was wrong about where the world was heading. If anything, it may have been right about the destination and wrong about the economics required to get there.</p><p>After leaving Wam!Net, a bunch of us joined a startup called Dantis. The company’s vision would later become known as Infrastructure-as-a-Service. Today, cloud infrastructure feels so obvious that it is difficult to imagine modern software development without it. At the time, however, it was still an emerging concept, and the surrounding ecosystem was far less mature.</p><p>When the dot-com bubble burst, raising capital became extraordinarily difficult. Dantis ultimately failed to secure additional funding and did not survive long enough to see the market develop. Looking back, I still find myself wondering whether we were wrong or simply too early. The distinction is easy to debate decades later, but it was much harder to see when the company depended on getting it right. One thing that experience taught me is that a good idea at the wrong moment can look remarkably similar to a bad idea.</p><p>Years later, I found myself riding another technology wave at Zoomdata. The promise of big data was compelling. Organizations were generating unprecedented amounts of information, new data architectures were emerging, and traditional analytics platforms were struggling to keep pace. Zoomdata built innovative technology designed specifically for that environment and attracted significant investment along the way.</p><p>Again, the broader trend proved real. Data volumes exploded. Cloud data platforms became mainstream. Analytics became a strategic capability for organizations of every size. Yet despite identifying an important trend and building innovative technology, Zoomdata ultimately sold for a fraction of its previous valuation. Other companies were able to acquire technology, talent, and capabilities at a much more attractive price than earlier investors had paid.</p><p>That experience reinforced something I had started to suspect from earlier chapters of my career. Being early and being right are not the same thing as capturing the value that gets created. I’ve watched waves of technology create enormous opportunities, but the companies that first identify them are not always the ones that benefit most. Sometimes the winners arrive later. Sometimes they have a different business model. Sometimes they simply take advantage of investments someone else has already made.</p><p>As I look back across these experiences, I don’t think there was a single lesson. The details were different each time. Architecture mattered more than I appreciated. Economics mattered more than I expected. Timing proved more important than I understood. What feels consistent, however, is that the underlying technology trend was often easier to predict than the path by which value would ultimately be created and captured.</p><p>That is why discussions around AI are so interesting. The conversation feels very familiar.</p><p>The early discussion focused on capabilities. Could large language models reason? Could they write code? Could they automate work? Will AI reshape knowledge work? Will it eliminate large categories of white-collar jobs? Will it create new kinds of work that we cannot yet envision? Will it justify the enormous investments being made in models, infrastructure, and applications?</p><p>More recently, the discussion has begun to shift toward economics. Companies are trying to understand token consumption, infrastructure costs, return on investment, governance, and pricing. Product leaders are debating how to package AI capabilities. Investors are debating where durable value will reside. Organizations are beginning to separate impressive demonstrations from sustainable business outcomes.</p><p>The conversation seems to follow a familiar pattern. It starts with “Can the technology do this?” Then people begin asking what it costs and whether the economics work. Eventually, the discussion shifts to who is actually capturing the value being created. That’s the question I keep coming back to.</p><p>One reason it is so difficult is that, with the potential exception of early adopters, customers rarely buy technology for its own sake (I do own an Apple Vision Pro, so I’m probably in that camp). They buy solutions to problems. Throughout my career, I’ve watched technologies create extraordinary possibilities while the companies that captured the most value were often the ones that successfully connected those capabilities to important customer workflows. The best technology does not always win. The winner is the company that makes the technology useful, accessible, and indispensable for solving a problem customers are willing to pay for.</p><p>The experiences I’ve described don’t make me skeptical about the technology itself. If anything, they’ve made me more cautious about assuming we know how the story ends. I suspect AI will prove every bit as important as advocates believe. Where will the value ultimately accumulate? With model providers, infrastructure companies, application vendors, or enterprises that successfully embed AI into their workflows, or companies that have not yet been founded?</p><p>History suggests the answer is rarely obvious in the middle of a technology wave. Having lived through several of them, I’ve become less interested in predicting whether a technology will matter and more interested in understanding how value will be created, captured, and distributed as it matures.</p><p>In my earlier article, I argued that successful technologies eventually become infrastructure. Looking back across my career, there is another lesson hidden in that transition. Long before a technology becomes infrastructure, people have to make bets about architecture, economics, timing, and where value will ultimately accumulate. Those decisions often determine who struggles, who succeeds, and who ultimately captures the value created when the future finally arrives.</p><p>Technology is hard. Some of it is truly amazing. Anyone who has spent time building products, platforms, or infrastructure knows that. The harder challenge is figuring out where the value will ultimately land.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0f9419d26d10" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Hardest Problem in Outcome Pricing Isn’t Measurement]]></title>
            <link>https://medium.com/@oliviersmeyer/the-hardest-problem-in-outcome-pricing-isnt-measurement-a06f3582d4d0?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/a06f3582d4d0</guid>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[product-packaging]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[pricing-model]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Fri, 29 May 2026 14:29:06 GMT</pubDate>
            <atom:updated>2026-05-29T14:29:06.725Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been following the recent discussion around AI pricing with interest. Much of the debate focuses on whether AI should be priced per user, based on usage, according to outcomes, or as a separate product capability altogether.</p><p>The topic resonates because I’ve had to make similar decisions across multiple products, with very different customers, cost structures, and usage patterns.</p><p>The assumption behind many of these conversations is that outcome pricing is the natural destination. If software can perform work directly, why not charge for the value delivered?</p><p>In some cases, that works remarkably well. Customer support is the obvious example. Products like Intercom Fin have demonstrated that customers are willing to pay for outcomes rather than seats or usage when the outcome is sufficiently clear and both parties can agree on what success looks like.</p><p>The farther you move from those types of workflows, however, the more complicated the picture becomes.</p><p>Consider Excel. It is one of the most successful software products ever created (I may be a little biased) and is used for thousands of different purposes.</p><p>If you wanted to implement outcome pricing, what would the outcome even be? Revenue generated, hours saved, better decisions, or errors avoided?</p><p>Even if you could measure those things, how much of the outcome should be attributed to Excel itself? A spreadsheet may contribute to a better decision, but so do the employee using it, the quality of the underlying data, the business process surrounding it, and the manager ultimately making the call.</p><p>The discussion reminds me a little of activity-based costing from my MBA accounting classes (many years ago now). The basic idea was that traditional allocation methods often didn’t accurately reflect what was actually consuming resources, so companies looked for cost drivers that better represented the underlying activities.</p><p>In some ways, AI pricing feels like a modern version of the same search. Seats, tokens, agent runs, resolutions, and outcomes are all attempts to find a unit that better reflects cost, consumption, or value. The challenge is that no single unit captures all three perfectly.</p><p>The more I think about it, the more I believe outcome pricing works best when three conditions exist:</p><p>1. The unit of work is discrete.<br>2. Success can be observed with relatively little disagreement.<br>3. There are few enough other contributors that customers can reasonably attribute the outcome to the software.</p><p>Customer support often meets those conditions. Many other categories do not.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EjIJc6kfDotzPxmDE1z_CQ.png" /><figcaption>Not all outcomes are equally suitable for outcome pricing.</figcaption></figure><p>The farther a product moves from discrete, observable work with few competing contributors, the harder outcome pricing becomes.</p><p>I’ve encountered this problem in various forms across different software businesses.</p><p>At Vbrick, we introduced AI capabilities such as transcription, translation, and facial recognition. Because those capabilities had different costs and usage patterns, we ultimately adopted a credit-based approach. Outcome pricing occasionally came up in the discussion, but attributing outcomes such as engagement, attendance, or content consumption to the platform itself proved challenging. The platform clearly influenced those outcomes, but it did not fully control them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0CddC2Pmmehy1M2HIahOog.png" /></figure><p>Years later at PerformYard, the challenge looked different but felt familiar. AI capabilities were packaged within a broader Premium offering alongside other advanced functionality and enterprise features. Outcome pricing sounds attractive in performance management software, but employee engagement and performance are influenced by managers, compensation, leadership, culture, and many other factors beyond the software itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Evz7hGIWJRkDiv_UXUxOtg.png" /></figure><p>More recently, while building GeoSheet, I found myself balancing a different set of tradeoffs. The base subscription includes a certain amount of AI usage, while heavier usage consumes additional credits. The goal was to allow customers to experience the value of the capability without requiring a separate purchase decision while still protecting the business from highly variable consumption costs.</p><p>One recurring tension across all three examples was adoption versus monetization. Separately pricing AI often made the economics easier to manage, but it also increased the likelihood that customers would never experience the capability in the first place.</p><p>These are very different products serving very different markets, yet they all surfaced similar questions. How do you balance customer predictability against variable costs? How do you encourage adoption without giving away unlimited usage? How do you price around outcomes when attribution is unclear?</p><p>The answers differed in each case, but none ultimately landed on pure outcome pricing.</p><p>At this point, someone will inevitably point to Intercom Fin or similar products. And they should.</p><p>Fin is one of the strongest examples of outcome pricing currently in the market. Customers are charged based on successful resolutions rather than seats or usage.</p><p>What’s interesting, however, is that Fin doesn’t necessarily solve attribution. Instead, it works because the conditions are favorable. A support interaction is a relatively discrete unit of work. Resolution can usually be observed. And while there are still multiple contributors to success, the number of plausible contributors is much smaller than in many other software categories.</p><p>Even then, the definition of success isn’t perfect. What counts as a resolution has required significant effort to define, measure, and validate, and those definitions continue to evolve. The point isn’t that attribution has been solved. It’s that both buyer and seller can agree on a practical proxy that is good enough.</p><p>The companies succeeding with outcome pricing are generally not proving causation. They’re finding outcome proxies that both buyer and seller are willing to accept.</p><p>The farther the software sits from the ultimate business outcome, the harder it becomes.</p><p>Excel struggles because the outcome is rarely discrete and almost always has multiple contributors. Performance management software faces a similar challenge. Even if employee engagement can be measured, managers, compensation, leadership, culture, and career opportunities all contribute to the result.</p><p>Video platforms sit somewhere in the middle. Attendance and content consumption can be observed, but the platform is only one factor among the speaker, the content itself, the audience, and the importance of the topic being discussed. The challenge isn’t that the outcomes can’t be measured. It’s that the software is only one contributor to the result.</p><p>This experience has made me skeptical of arguments that any single pricing model will eventually dominate.</p><p>Per-user pricing remains popular because it is simple and predictable. Consumption pricing better aligns costs with usage, but it introduces budgeting uncertainty. Outcome pricing can work when the conditions are right, but those conditions do not exist everywhere.</p><p>There are also adoption considerations. If AI is completely separated from the core product experience, customers may never discover its value. Product teams often want customers to experience new capabilities before asking them to make additional purchasing decisions. That reality has led many companies to include AI capabilities in subscriptions, provide limited usage allowances, bundle AI into broader product tiers, or adopt credit-based approaches that sit somewhere between subscriptions and consumption.</p><p>That may be why hybrid pricing models continue to emerge across the industry. Companies are trying to balance customer predictability, vendor economics, variable AI costs, product adoption, and perceived fairness simultaneously. In that context, hybrid pricing starts to look less like a temporary compromise and more like a practical response to competing constraints.</p><p>Today’s market encourages companies to make AI highly visible. Customers ask about it. Buyers expect it. Marketing teams highlight it.</p><p>At the same time, public attitudes toward AI appear to be becoming more complicated. Concerns about job displacement, layoffs, misinformation, and unrealistic expectations are creating pressure in the opposite direction. As AI becomes more commonplace, customers may become less interested in whether a product uses AI and more interested in whether it solves a meaningful problem.</p><p>If that happens, AI may follow the path of many successful technologies before it: becoming part of the product rather than a feature in its own right.</p><p>Attribution challenges make pure outcome pricing difficult. The gradual embedding of AI into products may eventually make separate AI pricing difficult as well. Both forces push toward broader packaging and hybrid models.</p><p>The current pricing debate often focuses on what should be measured: usage, outcomes, or value.</p><p>Those questions matter, but after working through versions of this problem across multiple products, I’ve become more interested in a different one.</p><p>Can buyer and seller agree on what success looks like?</p><p>In the narrow cases where they can, outcome pricing can work remarkably well. In many others, the challenge isn’t measurement at all. It’s attribution, agreement, and the reality that software is often only one contributor to the result.</p><p>That may be why so many companies keep arriving at some variation of hybrid pricing. Not because they’ve failed to find the “right” model, but because they’re trying to balance multiple competing realities at the same time.</p><p>It may also explain why, despite very different products, customers, and markets, I ended up wrestling with many of the same tradeoffs each time.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a06f3582d4d0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Minitel to AI: Living Through Seven Technology Shifts]]></title>
            <link>https://medium.com/@oliviersmeyer/from-minitel-to-ai-living-through-seven-technology-shifts-9a1e8de536ec?source=rss-f606c8886a5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a1e8de536ec</guid>
            <category><![CDATA[tech-history]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[station-f]]></category>
            <category><![CDATA[minitel]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Olivier Meyer]]></dc:creator>
            <pubDate>Wed, 27 May 2026 15:19:45 GMT</pubDate>
            <atom:updated>2026-05-27T15:19:45.088Z</atom:updated>
            <content:encoded><![CDATA[<p><em>They all followed a similar pattern (at least so far).</em></p><p>Last week, I was in France to attend a family funeral and had an opportunity to catch up with someone I had worked with 34 years ago.</p><p>Back then, we were trying to bring Minitel to the United States. For anyone who doesn’t remember it, Minitel was a combination of a small in-home terminal and an online service that existed before the web became mainstream. Messaging, chat, dating, directories, commerce, and information services all existed there in some form. Looking back, parts of it were surprisingly ahead of their time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*Z8AkwKkFtQ8uA3H8.jpg" /><figcaption><em>Minitel: messaging, chat, commerce, and online services before the web became mainstream.</em></figcaption></figure><p>Later that week, I found myself at Station F in Paris, a startup campus built inside a converted train station. There were founders, investors, startups, and larger technology companies all mixed together.</p><p>I decided to walk there rather than take the metro, from where I was staying in the 2nd arrondissement to Station F in the 13th. Since it was one of the many (many) long holiday weekends in France during May, or what the French call “les ponts de mai,” things were quieter than usual.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gPayqeVMN_-6qVA1VLKVUQ.jpeg" /><figcaption><em>Sometimes a few hours walking across Paris gives you time to think.</em></figcaption></figure><p>Walking around, I kept thinking about how much had changed. Thirty-four years ago, we were taking our own shot at bringing Minitel to the US. Now I was standing in a place full of startups trying to build the next thing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Fvm33p_qAgrUHuSaEJRUEg.jpeg" /><figcaption>Mix of startups and established tech companies.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*msXLpJPqpa_ELSdX51NVRA.jpeg" /><figcaption>The “food court” at Station F, all tech enabled, restaurants housed in the trains.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/768/1*O5KnNcz9gysqiwOsyT8ZVQ.jpeg" /><figcaption>Front entrance and welcome center for Station F.</figcaption></figure><p>It made me realize that I’ve now had the chance to watch several major technology shifts happen up close: personal computers, client/server systems, the web, SaaS, cloud, mobile, and now AI.</p><p>At the time, every one of them felt completely different.</p><p>Looking back, I think they may have followed a surprisingly similar pattern.</p><h3>Every Wave Feels Different While You’re Living Through It</h3><p>Each shift changed something meaningful. Personal computers changed who had access to computing. The web changed how people and systems connected. Cloud changed how infrastructure was delivered. Mobile changed where computing happened. AI may change the relationship between people and software itself.</p><p>The technologies changed significantly each time, which is probably why each wave felt unique while living through it.</p><p>But the patterns surrounding them often looked more familiar than I realized at the time.</p><h3>The Pattern Seems to Repeat</h3><p>Something new becomes possible, and the early versions are usually rough. Some people become convinced it changes everything, while others become convinced it changes nothing.</p><p>Then money and excitement start pouring in. Expectations get ahead of reality. Sometimes bubbles form.</p><p>Eventually, reality catches up. Infrastructure matures. Real use cases emerge. Some ideas disappear while others survive and become much larger than anyone expected.</p><p>Parts of this should sound familiar because Gartner formalized pieces of it through the Hype Cycle. Roy Amara captured another part years ago with the observation that we tend to overestimate the impact of technology in the short term while underestimating it over longer periods.</p><p>I think the pattern goes beyond technology itself. It seems to happen as much around people and behavior as around technology.</p><h3>They Don’t Just Repeat. They Build on Each Other.</h3><p>One thing I hadn’t really appreciated while living through these shifts is that they rarely replace what came before them. They usually build on top of it.</p><p>Personal computers put computing in individuals&#39; hands. Networks and client/server systems connected those computers. The web standardized access and distribution. Cloud made infrastructure elastic and accessible. Mobile put computing into people’s pockets.</p><p>AI is now arriving on top of decades of software, infrastructure, connectivity, and digitized workflows.</p><p>A smartphone without cloud services, APIs, app stores, GPS, and broadband connectivity would have been a very different experience. AI without large amounts of data, scalable infrastructure, and software embedded throughout businesses would likely look very different, too.</p><p>We often talk about technology revolutions as replacement events. Looking back, many of them feel more like layers.</p><p>Each one creates conditions that make the next one possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gojw73c4JUQRu1QsTRf1kw.png" /></figure><h3>Eventually, the Technology Disappears</h3><p>This may sound backward, but the technologies that become most important often stop being talked about as technologies.</p><p>Nobody says “internet shopping” anymore. It’s just shopping. We don’t think about whether email runs in the cloud or whether ride sharing started on a smartphone.</p><p>The technology fades into the experience itself.</p><p>The computer scientist Mark Weiser described something similar in 1991, arguing that the most profound technologies are the ones that disappear into everyday life.</p><p>Looking back, that feels increasingly right.</p><p>You stop thinking about the technology because it simply becomes part of how things work.</p><h3>Why I Think AI May Follow the Same Path</h3><p>Years ago, when I worked in BI and analytics, we used to make the case that analytics should appear within the workflows and applications people already used rather than existing as standalone tools.</p><p>The argument was fairly simple. Most people do not wake up in the morning wanting analytics. They want to hire employees, approve budgets, close deals, build products, or solve problems.</p><p>Analytics becomes valuable when it appears inside the work itself.</p><p>I increasingly see AI through a similar lens.</p><p>Today, AI feels like its own category. You see it everywhere: AI-native companies, AI-native products, AI-native applications, AI-native PMs, AI copilots, and AI assistants.</p><p>That probably makes sense right now because we are still early.</p><p>It’s possible this one is genuinely different. Plenty of very smart people think AI breaks the pattern rather than follows it. Still, the pattern has repeated often enough that I’d bet on it rhyming again.</p><p>But as AI becomes more embedded, I suspect it eventually stops being “the thing.” Instead, it becomes part of the roles, workflows, and applications people already use.</p><p>People ultimately won’t care whether something is AI-enabled. In some cases, they may not even want to think about AI at all. They’ll care whether it helps them do something better, faster, or more effectively.</p><h3>Back to Station F</h3><p>Walking around Station F, I was in a place built around people building what they believe may become the next generation of products and companies.</p><p>Some of those ideas won’t survive. Some companies won’t survive. Some of today’s assumptions probably won’t survive either.</p><p>But if previous technology revolutions are any guide, parts of today’s AI wave will eventually become ordinary.</p><p>And maybe that’s when you know a technology revolution has really happened.</p><p>Not when everyone is talking about it.</p><p>When nobody needs to anymore.</p><p>For those who are curious, here is a link to the Station F website: <a href="https://stationf.co/">https://stationf.co/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a1e8de536ec" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>