Centaurs Leading the Way to the Digital Cottage Industry
The real AI disruption may not be what you think it is.
There's a narrative circulating right now that goes something like this: AI is going to replace software developers. Vibe coding, agentic development, autonomous code generation — the machines are coming for the engineers.
I've seen this take in dozens of posts. Colleagues have sent me articles about it. And I think it's wrong — or at least, it's aimed at the wrong target.
AI on its own is not particularly good at building software. Not yet, and maybe not for a while. It hallucinates. It loses context. It makes architectural decisions that no experienced developer would accept. The idea that you can point an AI agent at a problem and walk away while it ships production-ready code is, for now, a fantasy.
But that doesn't mean AI isn't about to reshape the entire software industry. It absolutely is. The disruption just isn't coming from where most people expect.
It's coming from centaurs.
The Centaur Model
The term comes from chess. After Garry Kasparov lost to IBM's Deep Blue in 1997, he proposed a new kind of competition: human-computer teams, which he called "centaurs." The surprising result was that centaur teams — even average players paired with average computers — consistently outperformed both the best humans and the best computers playing alone. The combination was greater than either part.
In software development, a centaur is a skilled generalist who works with AI rather than being replaced by it. Someone with broad technical knowledge, product sensibility, design instincts, and enough domain expertise to ask the right questions and evaluate the answers. The centaur doesn't write every line of code — but they direct, evaluate, correct, and integrate the AI's output with the judgment that only experience provides.
I am a centaur. And I believe centaurs are about to make very large teams of specialized software professionals obsolete.
The Real Cost of Software
Here's something that most people outside the industry don't fully appreciate: the hard part of building software isn't the code.
There is almost no technical problem in enterprise software that can't be solved with some combination of data storage, retrieval, processing, and display. The fundamental operations are well-understood. The engineering, while sometimes complex, is rarely the bottleneck.
What takes all the time — what consumes the vast majority of the budget and energy — is the coordination of humans.
In my former role in enterprise software delivery, building a single product involved roughly a hundred people: executives setting direction, managers translating strategy into requirements, product managers defining features, project managers tracking timelines, scrum masters running ceremonies, architects designing infrastructure, backend developers building services, frontend developers building interfaces, designers creating user experiences, QA engineers testing everything, account managers liaising with clients, and the clients themselves providing feedback that then had to filter back through the entire chain.
Each of those roles exists not because the technical work requires it, but because the coordination requires it. The designer can't read the architect's mind. The product manager can't evaluate the engineer's implementation. The client can't speak directly to the developer without a translator. Every handoff introduces delay, misunderstanding, and cost.
Now consider what happens when one person — a centaur — can fulfill enough of those roles to collapse the coordination overhead almost entirely.
The centaur understands the business domain. They can talk to the client directly. They have the design sensibility to create an interface. They have the architectural knowledge to make sound infrastructure decisions. They have the technical skill to evaluate and direct the AI's code output. And they have the product instinct to know when something is good enough to ship.
One person and an AI. Doing what used to require a hundred.
The Enterprise SaaS Problem
This has profound implications for how organizations buy and use software.
Think about what it means to implement Salesforce, or SAP, or Workday, or Jira at an enterprise scale. These are sprawling, complex platforms that require dozens of specialized administrators, consultants, and trainers inside the organization just to manage them. Employees collectively spend hundreds or thousands of hours learning to operate tools that were designed for everyone and therefore fit no one perfectly.
The dirty secret of enterprise SaaS is that organizations don't adopt these tools because they're the best fit. They adopt them because building a custom alternative used to require a full development team — which meant a budget of hundreds of thousands of dollars, months of development time, and ongoing maintenance costs that only large organizations could absorb.
So instead, the organization reshapes itself to fit the software. You change your processes to match Jira's workflow model. You restructure your data to match Salesforce's schema. You train your people to think the way the tool thinks, rather than the other way around.
What if that calculus changed?
What if a centaur — one experienced technologist working with AI — could build you a bespoke tool that fit exactly how your organization was structured, reflected exactly what you built, and worked exactly the way your people already worked? A tool that required a fraction of the training time, none of the configuration overhead, and no annual licensing fee scaling into six or seven figures?
That's not hypothetical. That's what's becoming possible right now.
Why Big Companies Can't Just Do This
The obvious objection is: can't the big SaaS companies just hire a bunch of centaurs and offer the same thing?
They can try. But they'll run into a structural problem that's almost poetic in its irony.
The centaur model works because it eliminates coordination overhead. One person holds the full context — the business domain, the technical architecture, the user needs, the implementation details — in their head. The moment you try to scale that by hiring dozens of centaurs inside a large organization, you need to coordinate the centaurs. Which means you need centaur wranglers. And eventually someone needs to wrangle the wranglers.
You've recreated the coordination problem you were trying to solve. The large organization's structural disadvantage isn't a bug — it's inherent to being large. The centaur model is powerful precisely because it doesn't scale the way enterprises want things to scale. It scales by multiplication of independent operators, not by growing a single organism.
This is, incidentally, exactly how cottage industries have always worked. Not one factory producing a million identical widgets, but a thousand workshops each producing something fitted to their community's specific needs.
The Digital Cottage
This is the vision of the digital cottage industry.
Not a world where everyone builds their own software — most people don't want to and shouldn't have to. But a world where small, skilled operators build bespoke solutions for the communities and organizations they serve. Where software is crafted to fit, not forced to conform. Where the relationship between builder and user is direct, human, and ongoing.
Instead of a billion-dollar SaaS provider selling a one-size-fits-all platform to a million organizations, imagine thousands of centaurs each serving a handful of clients with tools built specifically for them. The economics work because the AI dramatically reduces the labor required. The quality works because the centaur actually understands the client's business. The sustainability works because the centaur isn't trying to achieve venture-scale growth — just a good living doing meaningful work.
That's what I'm building with sistnt. Not a platform for millions, but a small workshop producing carefully crafted tools for the people who need them. Every app I ship is proof that this model works — that one person with the right skills and the right tools can produce software that serves real needs without a hundred-person team behind it.
The Dark Cloud
I want to be honest about something, because I think intellectual honesty is what separates a genuine movement from a sales pitch.
This window may not stay open forever.
Every time a centaur works with an AI — directing it, correcting it, making judgment calls about architecture and design and user experience — that interaction generates training data. The AI learns not just from the code it produces, but from the corrections the centaur makes, the decisions the centaur explains, the taste the centaur applies.
If enough centaurs provide enough of this training data over enough time, it's plausible — maybe likely — that within a few years, the AI providers will be able to emulate the centaur's judgment well enough to cut the human out of the equation. At that point, the AI providers could price their agents beyond the reach of independent operators, and only large companies with deep pockets would be able to afford the bespoke, AI-driven development model.
I don't have a clean answer to this. Maybe competition between AI providers keeps prices accessible. Maybe regulation prevents monopolistic pricing. Maybe open-source models advance far enough to provide a viable alternative. Maybe the centaur's judgment proves harder to replicate than we think. These are genuinely open questions, and I intend to explore them.
But here's what I know: the window is open right now. The tools exist. The economics work. The skills can be developed. And every day spent waiting for certainty is a day not spent building.
The Call
If you're technically skilled, start broadening. Learn design. Learn product thinking. Learn how to talk to clients. The centaur isn't a specialist — they're a skilled generalist who can hold the full picture.
If you're a generalist, start getting more technical. You don't need a computer science degree. You need enough technical fluency to evaluate AI output, catch mistakes, and make architectural decisions.
And everyone — regardless of where you are — needs to start working with AI now. Not as a novelty, not as a toy, but as a collaborator. Because the centaurs who build their skills during this window will be the ones who establish the digital cottage industries that could reshape how software gets made.
The future of software isn't a factory. It's a village of workshops.
Start building yours.
George Pechtol is the founder of sistnt, where he builds privacy-first iOS apps to demonstrate that the digital cottage industry model works.
Published