--- title: "Clayton Christensen\'s Dream: Disruptive Innovation, as Economic Sharks Become Infrastructure" date: 2026-04-22 description: "Large companies are launching AI that serves their own ecosystems, not free innovation. Clayton Christensen explained the mechanism in 1997. An article about the time windows that giants leave open and what small companies can do about them." author: "Petru Cojocaru" categories: - name: "Science and civilization" url: "https://cojocarupetru.info/science-innovation-and-civilization.md" --- # Clayton Christensen's Dream: Disruptive Innovation, as Economic Sharks Become Infrastructure **At the PC, at home, with an idea that refuses to come out.** The cursor blinks. Blinks again. I've been staring at it for about ten minutes without having written anything — or rather, I've written and deleted the same first sentence three times, each time sounding either too grave, or too breezy, or simply false — that particular kind of false you sense immediately when you're writing not because you're thinking, but because you want to appear to be thinking. [![Clayton Christensen's Dream: Disruptive Innovation, as Economic Sharks Become Infrastructure](https://cojocarupetru.info/images/banners/inovatia_disruptiva.jpg "Clayton Christensen's Dream: Disruptive Innovation, as Economic Sharks Become Infrastructure")](https://cojocarupetru.info/index.php?option=com_content&view=article&id=316:clayton-christensen-s-dream-disruptive-innovation-as-economic-sharks-become-infrastructure&catid=37:science-innovation-and-civilization&lang=en-GB) Outside it's morning in Botoșani, which means cars can be heard on the street out front, the neighbour from the second floor has already switched on the television, and Gri is sitting on the windowsill watching a bird he will never catch, with a concentration I cannot replicate at the keyboard. I've had one coffee. I opened a second tab with Google News and closed it immediately, irritated by what was there — AI news, again, constantly, as though AI were the only thing that had happened in the world since 2022. I opened the file again. The cursor blinks. The idea I'm trying to write does exist — I'm certain of that. I can feel it somewhere in the region of my stomach, or perhaps higher up, in that imprecise place in the chest where things collect when you know them but haven't yet put them into words. It's connected to something I've been noticing over the past few months without being able to articulate clearly: **the major platforms have each launched their own AI, and all of these AIs resemble walls more than doors**. Google has launched Gemini integrated into Search, Gmail and Drive; Microsoft has embedded Copilot into Windows and Office; Meta has AI in Instagram and WhatsApp; Apple has announced Apple Intelligence across every native application. And when I look at all of this together, I don't see a revolution — I see large companies that have grown frightened and built defences. But I don't know how to put that in writing without sounding like a YouTube commentator who has read a thread on Twitter and fancies he has a theory. That's the problem. Not the absence of the idea — the absence of the architecture within which the idea can stand upright. My father was a mechanic. He repaired engines in the workshop at the back of the house, and sometimes he'd take me along when I was small. I would watch him tighten a bolt and knew there was a precise reason why he tightened it exactly that much and no more — not because he ever explained it to me, but because you could see in his gesture that he knew exactly what he was doing. This morning I don't know what I'm doing. I'm tightening bolts without knowing which engine I'm working on. I have another coffee and think about my son, who has had a mobile phone for two years and uses Google as a reflex rather than as a tool — he types something, receives an answer, no longer checks the source, no longer follows the link, stays within Google. And now, with Gemini integrated into search, he doesn't even reach the external link any more — the answer is right there, generated, elegant, complete, and entirely theirs. Google's. The Google ecosystem's. Delivered via Google, processed by Google, returned to Google. I feel this is an important thread. But the thread won't become text. It remains a thread. I close the laptop. Not dramatically, not in anger — simply with the acknowledgement that nothing is going to come out of me at the desk this morning. Sometimes thoughts need air, not a screen. --- ## Executive Summary *For the reader who has three minutes and wants the essence* --- The major web platforms — Google, Meta, Microsoft, Apple — have over the past two years launched their own versions of AI integrated directly into their existing products. On the surface, this looks like innovation. In reality, it is defence. Each of these AIs is built with an invisible but structural constraint: it cannot be better than the business model of the platform that created it allows it to be. Gemini cannot be optimised to send you outside Google. Meta's AI cannot be built to tell you that you spend too much time on Instagram. Copilot will never encourage you to switch to Google Docs. These are partisan tools — built to keep you within the ecosystem, not to help you think freely. This mechanism is neither new nor accidental. Clayton Christensen described it in 1997 in *The Innovator's Dilemma*: successful companies fail precisely because they do everything correctly — they maximise profit on existing products, listen to current customers, and avoid risks that might cannibalise their revenues. Real innovation always comes from outside, from the place where nobody has anything to lose. The practical consequence is that the partisan AI of the giants leaves behind uncovered territories — specific needs that a global product, built to serve billions of users simultaneously, cannot serve without compromising its business model. These territories are the time windows for small companies and open-source projects. The windows are real. They are narrow. And they are closing — not through deliberate intervention, but through the inertia of the very success of those who opened them. **Three conclusions for three types of reader:** For the independent developer — do not compete with Google on its own territory. Occupy the territories that Google cannot serve without contradicting itself. Build quickly, because the window is measured in months. For the ordinary user — the free AI integrated into the platforms you already use is not neutral. It has interests. Knowing this doesn't mean you must refuse it — it means you can decide consciously when you use it and when you look elsewhere. For the parent with a teenager in the house — the generation growing up with partisan AI as the norm is the generation that will have to live with the consequences of not understanding this mechanism. The conversation about how the tools we use every day work is worth having early, and without it sounding like a lecture. *The full article follows. If you only have time for the summary, you already have the essence. If you have time for the rest as well, you'll find the whole mechanism — with Christensen, with Ionuț in the park, with Gri on the keyboard, and with the mechanic father who knew that **you don't leave your tools with someone else overnight**.* --- ## The Park Walk as a Method of Thinking The park in the town centre has its midday Wednesday population: elderly people on benches with hands on knees and gazes that follow nothing in particular, two mothers with prams who know each other and are talking about something I can't hear, a German shepherd on a lead pulling towards pigeons with a hope the pigeons serenely ignore. I walk along the main path with my hands in my pockets and with that idea in my chest, which has loosened a little in the fresh air. I think better when walking. I don't know whether it's something neurological or simply habit, but at the desk thoughts lock into circles, whereas when walking they unfold linearly — one thing leads to another, threads extend without breaking. My father never went for a walk to think — he thought whilst working, with his hands in the engine, and he used to say that if you sit and think without doing anything you become depressed. He may have been right. Perhaps I need an engine to repair. I spot Ionuț on a side bench, with a coffee in a plastic cup and his mobile in his hand. Ionuț works somewhere in accountancy; we've known each other for years and greet one another with that particular warmth specific to people who respect each other without being close friends. I stop. We exchange a few words about the weather, about how it's cooled down again, and then he says something that makes me stay longer than I'd intended. *"You know, Google Maps suggested a restaurant to me yesterday before I even searched for it? I opened the app and it was already there, recommended, with stars and everything. I hadn't searched for anything. It knew on its own."* He says this with a mixture of admiration and a vague discomfort he cannot name. I know that discomfort. It's the same one I've had for the past few days. *"Yes,"* I say. *"And in Gmail now it completes your emails before you've finished writing them. And in Drive it suggests which document to open. And in Search you don't click on links any more — you get the answer directly."* *"It's convenient,"* he says. *"It's convenient,"* I confirm. *"But it's still all theirs. You stay in Google."* He looks at me for a moment. *"Well, where else would you go?"* And that, precisely, is the question. Where else? What alternative exists? And why doesn't it exist — because nobody has built it, or because every time someone begins to build it, Google launches its own version integrated into something you already have on your phone? I walk on and meet Mircea, who works at the town hall in an IT department and has a pragmatic perspective on everything. I tell him what I discussed with Ionuț and he immediately raises an example I hadn't considered. *"You know how Yelp died? Google copied local reviews into its own search and put the Google Maps results above everything else. Legally. It simply used its dominant position in search to make Yelp irrelevant without even buying it."* *"And Snapchat?"* I add. *"Stories. Instagram copied them exactly. Feature by feature. And Snapchat didn't die, but it stopped growing."* We sit on a bench and spend about twenty minutes talking about things each of us had noticed separately: that Facebook copied TikTok with Reels, that Microsoft copied Google Docs with Office Online, that Amazon copied third-party sellers' products with its own brands, that Apple copied third-party apps directly into iOS. Not out of malice — with something colder than malice, with pure economic logic. And someone says — I no longer remember whether it was me or Mircea — that all of this is happening especially now, with AI, because AI is the first tool that can be integrated invisibly into everything that already exists, that doesn't need a separate application, that can be planted in every product you already have, so that you have no reason to step outside the ecosystem. I walk back home with something that is no longer a vague idea. It's a pattern. I don't yet know what to call the pattern, but I know it exists and that I've seen it from three different angles in the past two hours — mine, Ionuț's, Mircea's — and that all three of us described the same thing in different words. --- ## Returning Home and Checking I go in, disturb Gri who was asleep on the keyboard — a typical protest move: he relocates to the chair with an air suggesting I am the one in the wrong — and open the laptop with a different energy than this morning. I'm no longer trying to write. I'm searching. The first searches are chaotic, exploratory: *"big tech companies copying AI start-ups"*, *"Google copying competitor apps"*, *"why tech start-ups don't grow any more"*. The results are partial, journalistic; each article describes a piece of the picture without seeing the whole. I come across the term *kill zone* — the area around the major platform ecosystems where investors refuse to fund start-ups because the probability of being copied or acquired before reaching profitability is too great. It's the right term for what I and Mircea had been describing on the bench, but still not enough. I search more specifically: *"innovator's dilemma big tech AI"*. And there, in the third or fourth result, the name appears: **Clayton Christensen**. **The Innovator's Dilemma. 1997.** I sit for a moment with the cursor hovering over the link without clicking. I have that specific feeling — not of discovery, not of intellectual triumph — but of mildly embarrassing recognition. That what I thought was my original observation from 2025, formed over coffee and reinforced during a park walk, is in fact a theory published almost thirty years ago by a professor at Harvard Business School. That the pattern I saw in the partisan AI of the major platforms, he saw in the hard-disk industry, in American steelworks, and in motorcycle manufacturing. I click. I read for an hour. I read summaries, reviews, a scanned chapter on a website where it probably shouldn't be, interviews with Christensen from 2012 and 2018, a year before he died. And with every page, with every paragraph, the idea in my chest clarifies — not because Christensen explains it to me, but because he confirms that what I had felt this morning in front of the blinking cursor was real and verifiable and part of a mechanism he described with a precision I shall never achieve. My son comes into the room, asks if there's anything to eat, I tell him there's soup from yesterday in the fridge. He asks what I'm looking at with such concentration. I say: *"I'm reading about a man who understood why large companies cannot save themselves."* He glances for a second at the screen, says *"oh,"* and goes off to find his soup. Twelve years old. Correctly prioritised. --- ### What Christensen Said and Why It Is More Relevant Than Ever Clayton Christensen published *The Innovator's Dilemma* in 1997 and changed the way the business world spoke about failure. Not about the failure of bad companies — about the failure of good companies. About the mechanism by which a company that does everything correctly, that listens to customers, maximises profit, and takes rational decisions, nonetheless ends up being overtaken and replaced. The central argument is simple and devastating in equal measure: successful companies don't fail because they are bad. They fail because they are too good at what they do. Their internal systems, their incentive structures, their relationship with existing customers — all of these are perfectly calibrated for the world in which they succeeded, and for precisely this reason are incapable of responding to a new world that emerges from the side. Disruptive innovation — the term Christensen popularised, though he later acknowledged it had been widely abused — does not usually arrive with a better product. It arrives with a worse product: cheaper, simpler, serving a segment that the market leader deliberately ignores because it isn't profitable. The first small hard drives were technically inferior to the large ones. The first Honda motorcycles sold in America were toys compared to Harley-Davidson. Netflix was posting DVDs when Blockbuster operated thousands of physical stores with excellent margins. And the market leader sees the threat — that is the subtlety Christensen always emphasises. It's not that they don't see it. They do. But the internal calculations, the financial logic, the pressure from shareholders, the organisational inertia — all of these say that responding to a marginal threat, aimed at an unprofitable segment, with inferior technology, does not justify cannibalising one's own profitable product. And it is right in the short term. And it loses in the long term. Now apply this to 2025. Google earns approximately 75–80% of its revenue from advertising in search. The model is simple and extraordinarily profitable: the user searches for something, Google displays links and adverts, the user clicks, advertisers pay per click. Every element of this model depends on a single thing: that the user stays in search, that they click on results, that the process has enough friction to generate advertising exposure. A genuinely good AI would destroy the model. If Gemini were built to answer every question completely and correctly, without any need to click on anything, advertising revenue would collapse. So Gemini cannot be built that way. Not because Google's engineers can't — they can. But because no manager with quarterly financial targets will approve a product that destroys the company's primary source of revenue. So Gemini is built to defend. Not to innovate. It is a partisan AI — partisan to the Google ecosystem, partisan to the Google business model, partisan to the survival of Google in its current form. And precisely the same is true of Microsoft's Copilot, Meta's AI, and Apple Intelligence. They are all walls. Built to keep you inside, not to help you get out. Christensen would have recognised the pattern immediately. He saw the same mechanism in Sears, in US Steel, in Kodak. The companies are not bad — they are rational. And short-term rationality produces, systematically, long-term irrationality. --- ### The Gaps Giants Leave When They Defend Themselves There is a structural consequence of partisan AI that nobody discusses with sufficient seriousness: when you build a defensive product, you cannot build it well across the board. You inevitably make compromises. And these compromises are not accidental — they are necessary, inscribed in the logic of the product. An AI integrated into Google Search cannot be optimised to send you outside Google. That would be institutional suicide. So it will be systematically weaker in situations where the correct answer lies on an external site, in an academic database, in an independent publication. It will present information with a slight gravitational pull towards Google-friendly sources, towards content indexed in Google's services, towards answers that do not require you to leave the ecosystem. It is not a conspiracy. It is architecture. An AI integrated into Meta cannot be built to tell you that you spend too much time on Instagram. An AI integrated into Amazon cannot be built to direct you towards a cheaper product from a competitor. An AI integrated into Microsoft Office cannot be built to suggest switching to Google Docs when it would suit what you're doing better. Every partisan AI has blind spots. And these blind spots are a map for anyone who wants to build something small and real. Christensen called underserved territories *non-consumption* — people who didn't consume a product at all because it was too expensive, too complicated, or simply non-existent for their specific needs. Disruptive innovation doesn't usually replace an existing product — it creates a market that didn't exist before. The Sony Walkman didn't replace home hi-fi equipment. It created the market for mobile music, which had not previously existed. Where are the non-consumers of current AI? They are the doctors in district hospitals who need diagnostic support but cannot send patient data through American servers. They are small local councils that would benefit from AI for processing administrative documents but cannot adopt cloud solutions without compliance with Romanian legislation. They are pupils in rural areas who need individualised tutoring but don't have access to tools designed for them. They are solicitors, accountants, engineers who work with technical documentation in Romanian and whose specific needs have never been on the list of priorities of a product designed for the American market. None of these markets interests Google. Or Meta. Or Microsoft. Not because they are malicious — because these markets are small and fragmented and don't justify the effort of calibration for a product designed to serve billions of users with similar behaviours. These territories are the windows. Not metaphorically — literally: open, uncovered spaces in which a technically more modest product, perfectly calibrated for a specific need, can grow without being caught in the giants' kill zone, at least for a while. --- ### Why It Doesn't Eat Its Own Tail The question I receive every time I explain Christensen's mechanism to someone is a variant of the same puzzlement: *"But don't they see what's happening? Why don't they adapt?"* They do see. That is precisely the point. Imagine being a product manager at Google in 2024. You have a team of thirty engineers, a budget of several tens of millions, and product objectives tied to Search metrics — click-through rate, time on site, ad revenue per query. Someone comes to you with a proposal: *"Let's build an AI that answers questions so well that the user no longer needs to click on anything."* You, as manager, know that this proposal is correct in the long term. You know that this is the direction the world is heading. You know that if you don't do it, someone else will. But if you do it, click-through rate falls, ad revenue per query falls, your quarterly metrics decline, at performance review you are evaluated on the basis of these metrics, your bonus depends on them, your career within the company depends on them. What do you do? You build an AI that looks impressive in a demo, that answers simple questions, that retains enough friction to preserve click behaviour, and that is integrated tightly enough into the ecosystem to appear innovative without being dangerous to the business model. You build a partisan AI. You are not a coward. You are not obtuse. You are rational within a system of incentives that produces precisely this result. Christensen understood that the problem is not the people inside the company — it is the structure of the company. Resource allocation mechanisms, project approval processes, performance evaluation systems — all of these are built around the existing business and systematically reject any project that threatens it, even when that project would be correct in the long run. The solution Christensen proposed was to create a separate organisation, with its own incentives, its own culture, its own financial structure — detached sufficiently from the parent company to be able to innovate without being killed off by institutional antibodies. A few companies tried. Few succeeded. The separation is always incomplete, always polluted by the culture and politics of the parent company. Amazon Web Services is perhaps the best example of success — a business that at its outset would have cannibalised Amazon's retail model, separated sufficiently to grow, and which today generates more profit than all the rest of Amazon combined. But Bezos had to fight internally for years to protect it. How many Bezoses are there at Google? At Meta? At Apple? --- ### Windows Are Compressing — The Urgency of the Accelerated Cycle Christensen worked with examples from industries in which the disruption cycle was measured in decades. Kodak saw the digital camera in 1975 — they invented it themselves — and died in 2012. Blockbuster saw Netflix in 1999 and went bankrupt in 2010. Eleven years. Fifteen years. Time enough for a generation of managers to retire with comfortable pensions before the disaster reached them. In AI, the window is eighteen months. Perhaps twenty-four. The iteration pace of language models, the exponentially falling cost of training, the democratisation of access to cloud infrastructure, the availability of open-source models that can be fine-tuned with relatively modest resources — all of these compress the Christensen cycle to a speed that no longer leaves room for mistakes. A start-up that identifies a window in January 2025 and builds a product within that window has limited time before Google, Microsoft, or Meta notices the window exists and launches their own version, integrated into something the user already has. Not a technically better version — a version *good enough, delivered through a channel the start-up cannot replicate*. This is the real advantage of the giants. Not the technology. The channel. If Google launches a mediocre AI feature tomorrow directly in Gmail — which you already have open for eight hours a day — that mediocre feature will be used by more people than the best independent product that requires a separate application, a new account, a change of habit. Distribution beats quality in rapid scaling. Always. So the advice for anyone building within this window is not to build something technically better than what Google will launch — but to build something specific enough, deep enough in the need, sufficiently integrated into the context of your particular user, to survive the moment when Google launches the mediocre, generic equivalent. And to build quickly. --- ## Probable Predictions — What Christensen Would Say If He Were Alive in 2025 Christensen died in January 2020, a few months before GPT-3 changed the conversation about AI. He never saw ChatGPT, never saw Gemini, never saw what had become of the industry he had analysed with such precision. But he left an analytical mechanism robust enough that we can extrapolate what he would have said. **First prediction: the wave of acquisitions.** Large companies always buy what they cannot copy in time. But they buy too late, after the window has opened enough to validate that the innovation is real and the market exists. Christensen documented that the acquisition usually comes not because the product being acquired is profitable — but because its presence in the market is uncomfortable, confirms that there is an uncovered territory, and attracts other investors' attention towards that area. In AI, the second and third waves of acquisitions are coming. Perplexity, Mistral, Runway, companies building AI for jurisprudence, medicine, education — not all will be acquired, but enough to confirm that the giants continue to react rather than to anticipate. **Second prediction: open source as an uncapturable fortress.** Christensen wrote relatively little about non-commercial mechanisms of innovation — his theories were calibrated to companies with a profit logic. But had he applied the framework to open source, he would have observed that open-source models eliminate precisely the vector of defence that the giants deploy most effectively: control of the distribution channel. You cannot buy an MIT licence. You cannot force the disappearance of a model running on the servers of tens of thousands of people simultaneously. You can fork it, improve it, launch a commercial version — but you cannot stop the community from continuing in its own direction. Llama, Mistral, Phi — all the open-source models growing in performance with every passing month — represent a type of disruptive innovation that Christensen did not explicitly catalogue but which his framework predicts: innovation emerging from the non-commercial segment, for users who are not profitable for large companies, and growing precisely in the area where the giants cannot defend themselves without contradicting themselves. **Third prediction: innovation moves to non-consumption.** The most interesting thing Christensen said is not about companies that fail — it's about markets that are created. Disruptive innovation does not always replace an existing product. Sometimes it creates a market that didn't exist before, serving people who had no access to any prior product. Where are the non-consumers of current AI? They are the doctor at the district hospital in Vaslui who would benefit from diagnostic support but cannot use a tool that sends patient data to servers in California. They are the judge processing hundreds of case files who would benefit from an AI that summarises legal documents in Romanian, with Romanian jurisprudence, but no such tool exists because the market is too small for a global product. They are the teacher in a rural area who could use AI for individualised pupil preparation, but the existing tools are not calibrated for the Romanian curriculum. None of these markets interest Google. Or Microsoft. Or Anthropic. They are small, fragmented, dependent on local contexts that a global product cannot incorporate efficiently. But they are real. And they are uncovered precisely because they are too small for the giants and too specific for a generic product. **Fourth prediction: the giants will survive, but they will become infrastructure.** Christensen did not predict the death of market leaders — he predicted their transformation. IBM did not die when it lost the personal computer market. It survived by becoming a provider of enterprise solutions, a consultancy, an infrastructure provider. It won differently from how it had won before — less spectacularly, less visible in popular culture, but profitably and stably. Google, Meta, and Microsoft will not disappear. They will become the infrastructure on which others build; they will earn from cloud, from APIs, from data processing — not from final consumer products. The partisan AI they are currently building is the last mass consumer product they will truly own — what follows will be built by others, on their platform, and they will earn a percentage from each transaction. Christensen would have seen in this not a defeat — but a predictable, documented, inevitable transformation. **The agentic era, or the rush for illusory gold? A 2026 confirmation of Christensen's dilemma** I recently examined an idea that circulates insistently in technology circles and which, on first reading, appeared to contradict everything I've written above. Eric Schmidt, Google's former CEO, was telling start-up founders at Davos that "if you really want to make money, start an AI agents company." Everyone will build agents, he said, and these agents will compete fiercely. The opportunity is no longer in the large models, but in practical application, in the [specialised agent](https://edition.cnn.com/2026/03/25/us/word-of-week-agentic-ai-cec) that executes real tasks. I sat with this in front of me one whole morning, with a coffee that went cold too quickly, and with the feeling that I was hearing a call to a new digital gold rush. That anyone, from a garage or a flat in Botoșani, could build that agent that solves a real problem and becomes the next big winner. It's a seductive narrative, especially for someone who, like me, believes in the power of small, independent initiatives. The trouble is that it sounds too good. And when something sounds too good in technology, it usually conceals a mechanism you haven't yet seen. Schmidt did indeed predict an era of agents. But the idea that this represents a quick and easy route to wealth is [a dangerous oversimplification](https://asksurf.ai/pulse/zh/schmidt-%25e7%258b%25ac%25e7%25ab%258b%25e5%2588%259b%25e5%25a7%258b%25e4%25ba%25ba-%25e6%2599%25ba%25e8%2583%25bd%25e4%25bd%2593%25e6%2597%25b6%25e4%25bb%25a3-%25e8%25af%25af%25e5%2588%25a4%25e8%25a7%25a3%25e6%259e%2590). The reality, as more careful analyses reveal, is that the market will be extremely competitive, and the winners will most likely be those with massive resources and proprietary data — not any "weekend founder." Building a basic agent has already become a trivial exercise, almost a standard API call. The real value — and therefore the real barrier — lies in exclusive data, deep integration with the client, and governance that offers trust to large companies. And all of this is precisely the type of resource that the giants possess in abundance. And, to close the circle neatly, I've watched these giants move exactly as Christensen's framework predicts. Whilst small founders dream of revolutionary agents, Google is consolidating its position not by building every possible agent, but by becoming the infrastructure on which all these agents will be forced to run. Their recent acquisitions betray this strategy with an almost cynical clarity. On one hand, they acquired Wiz for 32 billion dollars. Wiz is not an AI agent, but a cloud security platform. For Google, which wants to become the playing field of the "agentic era," security is not a function — it's the foundation on which you build trust. It's the wall of the fortress they are promising corporate clients will keep their data and agents safe. On the other hand, they "acquired" the team behind Windsurf, an "agentic coding" tool, paying 2.4 billion dollars. This is no longer a brick — it's a tool. They are not merely buying another agent, but the very engine that will help others build and refine agents. In the logic of the gold rush, Google is positioning itself to be the one selling the best pickaxes and shovels — whilst also owning the map and charging a levy on every seam discovered. This is what makes Christensen so relevant in 2026. We are not watching a simple innovation race, but a predictable reaction from the "economic sharks," who, sensing a potential disruption, cannot confront it directly without cannibalising themselves. And so they choose a different path: they become the foundation on which the new is built. Their profit will not come from the next brilliant product, but from the fact that every brilliant product will need their infrastructure to exist. For an independent developer, this is neither a cause for despair nor a call to rapid enrichment. It is a reminder that the real opportunities are most likely in the niches this infrastructure ignores — and that the greatest trap is to play the game on someone else's pitch, by someone else's rules. The window is there, but it is not a wide gate. It is a crack that demands precision, not blind haste. --- ## Conclusion — The Blogger Who Went for a Walk and Came Back with a Law It's evening now. Gri has migrated from the keyboard to the armchair. My son is in his room with headphones on. Outside on the street in Botoșani the traffic has gone quiet. I wrote more today than I expected when I was stuck in front of the blinking cursor this morning. Not because I was more intelligent in the afternoon than in the morning — but because I went out, walked, talked with Ionuț and Mircea on a park bench, listened to an observation about Google Maps suggesting restaurants without being asked, and let myself be carried by that thread which already existed in my chest, unseen. My father used to say that there is no engine so broken that nobody can work out how to repair it — there are only mechanics who haven't walked around it enough times to understand what it is producing. This morning I stood in front of the engine and looked at it. On the walk I did the laps. By evening I understood what it was producing. Clayton Christensen published this understanding in 1997, when I was a teenager and the internet was a modem sound on a telephone line. He walked around the engines in the hard-disk industry and American steelworks and saw that they all produced the same sound — the sound of a successful company that cannot afford to destroy itself in order to rebuild. I am not original. That is the most honest conclusion of the day. What I saw in the integrated AI of Google, Meta, and Microsoft; what Ionuț felt when Maps suggested a restaurant without being asked; what Mircea articulated in the story of Yelp and Snapchat — all of these are 2025 examples of a mechanism described in 1997 by a man who never lived to see confirmed the last and most spectacular act of his theory. The cannibals don't appear from outside. They grow from within, from the rib of their own fear. And that, at the same time, is the best thing that could happen for anyone small enough to have nothing to lose. --- ## Bibliography — Online Sources Without a Subscription --- **Clayton Christensen — The Original Theory** - *The Innovator's Dilemma* — summary and academic context: [https://claytonchristensen.com/key-concepts/](https://claytonchristensen.com/key-concepts/) - Harvard Business Review — Christensen explains disruption in his own words (freely archived article): [https://hbr.org/2015/12/what-is-disruptive-innovation](https://hbr.org/2015/12/what-is-disruptive-innovation) --- **Kill Zone and the Acquisition Behaviour of the Giants** - The Economist — *The techlash against Amazon, Apple, Facebook and Google*: [https://www.economist.com/briefing/2018/01/20/the-techlash-against-amazon-apple-facebook-and-google](https://www.economist.com/briefing/2018/01/20/the-techlash-against-amazon-apple-facebook-and-google) - Vox — *Big Tech's kill zone*: [https://www.vox.com/recode/2019/6/18/18681572/big-tech-kill-zone-startup-google-amazon-facebook](https://www.vox.com/recode/2019/6/18/18681572/big-tech-kill-zone-startup-google-amazon-facebook) --- **Partisan AI and Integration into Existing Ecosystems** - The Verge — *Google is turning Search into a walled garden*: [https://www.theverge.com/2023/5/10/23717120/google-search-ai-answers-sge-overviews](https://www.theverge.com/2023/5/10/23717120/google-search-ai-answers-sge-overviews) - Wired — *The AI features built into everything are mostly defensive*: [https://www.wired.com/story/microsoft-google-ai-features-defensive/](https://www.wired.com/story/microsoft-google-ai-features-defensive/) --- **Open Source as a Disruptive Response** - MIT Technology Review — *The open-source AI boom*: [https://www.technologyreview.com/2023/05/12/1072950/open-source-ai-google-llama-mistral/](https://www.technologyreview.com/2023/05/12/1072950/open-source-ai-google-llama-mistral/) --- **Documented Historical Cases — Yelp, Snapchat, Netflix** - Wikipedia — *Disruptive innovation* (historical examples section, with primary sources linked): [https://en.wikipedia.org/wiki/Disruptive_innovation](https://en.wikipedia.org/wiki/Disruptive_innovation) --- *Practical note: links to HBR and The Economist may require a free account on first access — registration is without charge. The remainder are fully open.* --- ## 10 Questions and Answers About the Article's Usefulness **1. Who does this article specifically address?** Anyone who produces or intends to produce something digital — a website, a plugin, a service, a product based on AI — and feels that they are playing on a pitch they didn't choose. But also the ordinary reader who uses Google, Instagram, and ChatGPT and has a vague sense that something isn't quite right about the way all of these function together, without being able to name what exactly. The article is not written for specialists who already know Christensen — it's written for the person who has felt the pattern on their own skin, without yet having the vocabulary to describe it. **2. What does an independent web developer gain from reading it?** A mental framework — which is more valuable than any specific practical advice. A framework helps you assess new situations yourself, rather than applying old recipes to new problems. Concretely: you understand why it makes no sense to build a product that competes head-on with Google or Meta on their own territory; you understand that the structural blind spots of the major platforms are not accidental but architectural; and you understand that the time window in which a small product can exist without being copied or bought is real but limited, and demands urgency rather than perfectionism. **3. Is it of any use to someone who doesn't build digital products?** Yes, and perhaps for that very person it is most useful, because they come without professional defensiveness. The article explains an economic mechanism that governs which applications appear on your phone, which features you receive for free and why, why some online tools suddenly disappear or change unpleasantly, why adverts proliferate in products that initially had none. Understanding this mechanism doesn't necessarily change what you use, but it changes the way you understand your relationship with the platforms you use — and that is a form of digital literacy more honest than any tutorial about privacy settings. **4. Why the narrative structure — morning at the PC, the walk, the return — rather than directly the theory?** Because theory alone is dead on the page without flesh on its bones. Christensen himself understood this — which is why he used concrete industrial examples rather than abstractions. The narrative structure of this article mimics the real process by which an idea forms: not at a desk, in front of a deadline, but in chance conversations with people who don't know they have a theory, yet are living it. Ionuț saying that Maps suggested a restaurant without being asked is more persuasive than any academic citation, because Ionuț is real and his experience is verifiable by any reader who has their phone in their pocket. **5. What risks does a hurried reader misunderstand?** That the article is an anti-Google or anti-Big Tech manifesto, which it is not. It isn't even a moral critique — it is a description of an economic mechanism that produces predictable results regardless of the good or bad intentions of the companies involved. Google is not bad. Meta is not bad. They are rational within a system of incentives that systematically produces defensive innovation rather than open innovation. Christensen was explicit about this: the innovator's dilemma is not a problem of character, it is a problem of structure. The reader who comes away angry at Google has missed the central point. **6. How valid are the predictions in section eight?** As valid as any prediction based on mechanism rather than detail allows — that is to say, considerably more valid in direction and considerably less valid on timing and specific actors. That a wave of AI acquisitions in niche markets will follow is almost certain, because it is exactly what happened in every previous technology cycle. Which particular start-up will be bought, when, and at what price — nobody knows that. The value of the article's predictions is not that they tell you what to buy on the stock market, but that they describe a horizon of probability within which to calibrate your own decisions. **7. What is missing from the article that should be added in an extended version?** Missing are voices from within the large companies — the engineers and managers who know they are building defensive rather than innovative products, and who are themselves living Christensen's dilemma at a personal rather than institutional level. Also missing are more concrete local examples: local start-ups that built something real and were not killed through acquisition but through irrelevance — through the launch of a similar feature directly within the dominant platform. And missing is a section about what happens to the ordinary user in the long term — not the product builder, but the person who consumes partisan AI and, gradually, no longer knows that other territories exist. **8. Can the article be used as an argument with a client who doesn't understand why they should invest in their own digital product rather than rely on existing platforms?** Yes, and it is one of the best structural arguments for this conversation, because it does not appeal to diffuse fear but to demonstrable logic. The client who depends entirely on Google for visibility, on Meta for audience, on Microsoft for productivity — that client is, in Christensen's terminology, a non-consumer of their own digital independence. The window in which they can build something of their own exists now. In a few years' time, when the platforms are even more integrated and even more partisan, the window will have closed and the alternatives will be more expensive, more complicated, or non-existent. **9. What makes this article different from the dozens of texts about AI published daily?** It doesn't start from AI. It starts from the human — from the blinking cursor, from the coffee, from Ionuț on the bench, from Gri on the keyboard. AI is the subject, but it's not the starting point. The starting point is the feeling that something is wrong — a feeling millions of people have, and which very few texts validate before explaining. Most articles about AI make you feel you should inform yourself better, keep up, not fall behind. This article tells you that what you've already felt is correct, and that it has a name nearly thirty years old. **10. Does it merit translation and publication in English?** Yes, but not as a direct translation — as a rewrite for a different cultural context. Botoșani, the mechanic father, Ionuț from accounting, Gri on the windowsill — all of these work in Romanian because they are specific and credible within their context. Translated literally, they become local colour for a Western reader who doesn't feel them as his own. An English version would need to preserve the narrative structure and the analytical mechanism, but replace the personal anchors with others equally specific and credible for a reader from a different geography. The theory is universal. The flesh on it must be local. --- ## What Digital Behaviours Does the Article Suggest? *A practical section — not a manifesto, but a list of small habits with significant consequences* **1. Check the source of the answer, not just the answer** When Gemini, Copilot, or any integrated AI gives you a direct answer within the platform's interface, make a habit of asking: *where does this come from, and is there an external source I can verify?* Not out of paranoia, but as intellectual hygiene. The partisan answer is often correct — but it is correct in the direction of the ecosystem that produced it. An external source takes you outside the platform's editorial gravity and gives you a point of comparison. **2. Use at least two different search engines for important questions** Not because one is absolutely better than the other, but because each has different blind spots, different partisan interests, different ranking algorithms built for different ecosystems. Perplexity, DuckDuckGo, Kagi, Brave Search — none is neutral, but diversity reduces dependence on a single editorial filter. It is the digital equivalent of *don't read only one newspaper*. **3. Install at least one open-source AI model locally and use it periodically** Ollama with Mistral or Llama running locally on your own computer sends your data nowhere, has no product to defend, and has no business model to sustain. It is inferior in raw performance compared with GPT-4 or Gemini, but precisely this inferiority is instructive — it shows you what AI can do without the partisan constraints of a commercial platform, and calibrates your expectations more realistically than any polished demo from a major product. **4. When a platform launches a new free feature, ask what behaviour it is buying with it** Nothing is free within partisan digital ecosystems — it is an exchange of behaviour for service. Google gives you Gemini for free in Search in exchange for your usage data for model training and in exchange for keeping you within the Google ecosystem. Meta gives you AI in WhatsApp in exchange for access to your conversations as training data. You are not obliged to refuse the exchange — but you are obliged to know you are making an exchange, and to decide consciously whether it is fair to you. **5. Build an information flow that doesn't depend on a single aggregator** RSS still works. Direct newsletters from authors still work. Manually organised bookmarks still work. If everything you read reaches you through the filter of a single algorithm — whether Google Discover, the Facebook Feed, or Apple News — it means you are reading what a partisan system has decided is relevant to you, not what is actually relevant to you. The difference seems small. In the long run, it is the difference between a diverse information landscape and a well-upholstered echo chamber. **6. If you produce online content, publish on your own domain before any other platform** Whatever you write, record, or produce — if it exists first on your own domain, with your own metadata, with your own canonical URL, you are the owner of the first instance. If it exists first on Facebook, Medium, or LinkedIn and only thereafter on your own site, you have given first indexing rights to a partisan platform that can change its algorithm, restrict visibility, or disappear. My father used to say that you don't leave your tools with someone else overnight. The same principle. **7. Treat dependency on a single ecosystem as a business risk, not a convenience** If your business depends on Google Ads for traffic, Meta for audience, Amazon for distribution, or the Apple App Store for the existence of your product — you have a concentration risk you have probably never quantified. Christensen didn't write about start-ups, but the mechanism applies: the platform you depend on has its own partisan interests, which can diverge from yours at any moment. Diversifying channels is not paranoid perfectionism — it is elementary risk management. **8. Actively seek out products and tools built for the territories the giants ignore** There are AI tools for rural medicine, for jurisprudence in minor languages, for individualised education in non-American school systems, for local public administration — all built by small teams or open-source communities, all technically inferior to the products of the major platforms, all extraordinarily precise in the needs they serve. Using them when they are relevant to you is both an act of digital literacy and an economic vote for the kind of ecosystem in which you want to live online. **9. Explain to a teenager in your life what a partisan AI is** Not as a lesson, not as a lecture — as a conversation. My son is twelve and uses Google as a reflex. He doesn't know that the answer he receives is filtered through the economic interests of whoever produced it. Not because he is careless — because nobody has ever told him that there is a mechanism there, that the answer is not neutral, that the tool has its own interests. The generation growing up with partisan AI as the norm is the generation that will have to live with the consequences of not understanding this mechanism. That conversation is worth having early. **10. Read *The Innovator's Dilemma* — at least the first two chapters** Not because Christensen is infallible or beyond reproach — his theory has been criticised, amended, and contextualised across the twenty-eight years since it appeared. But because there is a rare category of books that give you an analytical mechanism you can apply to any industry, any market, any technological moment. The first two chapters of *The Innovator's Dilemma* are sufficient to understand the basic structure. The rest is exemplification. You will have the mechanism after fifty pages and will use it for the rest of your life, every time a large company appears not to understand why it is losing ground to a competitor it ignored for too long. ---