The One-Person Unicorn Won't Run on ChatGPT

Sam Altman and other leading AI researchers expect the first one-person unicorn — a billion-dollar company run by a single founder orchestrating AI agents — to appear somewhere between 2026 and 2028. Over roughly the same horizon, analysts put the enterprise AI applications market at around USD 1 trillion by 2030.

Set those two forecasts next to each other and an obvious question falls out. Will that one-person unicorn run on ChatGPT? Will the Fortune 500 spend a trillion dollars on Microsoft Copilot?

Obviously not. ChatGPT is a chat interface. Copilot is a fairly basic assistant bolted onto Office and Windows. Both are genuinely useful and both deliver real productivity gains, but they are the first step into the AI age, not the destination. Neither executes an end-to-end business workflow. Neither will absorb a trillion dollars of enterprise demand. Neither can run a company without staff.

Which leaves the question almost nobody is asking yet: what does the software actually look like that can deliver on these forecasts?

Nobody can answer that with certainty. But we are not reasoning in the dark either, because we have watched this happen before.

“What does the software actually look like that can deliver on these forecasts?”
Jörg Bröker

The internet, on two tracks

When the internet arrived, two things happened, and they happened separately.

The first was technical. In the beginning there were static, hand-written HTML pages. Then came JavaScript, AJAX and CSS, turning flat documents into interactive, well-designed websites. In parallel the first frameworks emerged, letting developers build on shared building blocks instead of reinventing the wheel for every new site. CMS and e-commerce systems followed. Today, modern full-stack frameworks cover frontend, backend and database, and let us assemble web applications in a fraction of the time — applications powerful enough to have replaced native Windows and Mac software. The infrastructure beneath made the same leap, from local servers to data centers to cloud computing and CDNs.

Each step did the same thing: it took a problem that every team had been solving on its own and turned it into something you could simply assume was there.

The track that mattered more

The second track is the striking one. Entirely new categories of business appeared on top of that infrastructure, and several of them became the largest corporations in the world.

Search, online advertising and cloud computing built Google. Social media changed the way we interact and built Facebook and Meta. E-commerce changed the way we shop and built Amazon and Shopify. Streaming changed how we consume media — and what content itself looks like — building Netflix, YouTube, Spotify and TikTok. Online payment built Stripe and PayPal, which are now encroaching on classical banks and payment providers. And SaaS rewrote the economics of software itself, turning a product you buy into a service you rent.

None of these categories existed before the internet. All of them are obvious in hindsight.

“None of these categories existed before the internet. All of them are obvious in hindsight.”
Jörg Bröker

That last part is worth sitting with. Nobody in 1995 was confidently describing the streaming industry. The categories were not visible from the starting line; they became visible as the stack underneath them matured and made them cheap enough to try.

So where are we standing?

The same evolution is now beginning with AI and AI agents. The software stack we work with today, and the business models built on top of it, will change just as radically. The trillion dollars will not be spent on chat windows. It will be spent on categories of software that mostly do not exist yet.

Which makes one question more useful than any prediction: where in that cycle are we right now?

“Where in that cycle are we right now?”
Jörg Bröker

The honest answer is somewhere around 1995. No standard stack. No shared building blocks. Every company solving the same hard problems from scratch, and hiring scarce, expensive experts to do it.

1995 was a very good year to start building...