The tiny-team, billion-dollar company already happened, and the conditions that made it possible are spreading

Elad Gil, the investor and entrepreneur, puts a sharp point on a claim that tends to get buried in speculation about AI futures: the tiny-team, multi-billion dollar company is not a hypothesis. Minecraft, he notes, was built by roughly five to ten people when Microsoft acquired it for billions. The deal is done. The precedent exists. The argument that such outcomes require AI to become possible is, on its face, wrong.

What has changed is not that tiny teams can now build enormous things. It is that the cost floor for attempting it has dropped, and the number of people in a position to try has expanded. Mark Cuban describes having two or three people building software that, five years ago, would have consumed two or three million dollars a year. The work is the same. The resource requirement is not.

Patrick Collison, co-founder of Stripe, offers a concrete internal example. An AI knowledge tool the company built internally took two people roughly six months to complete. Collison also notes that year-on-year usage growth of Stripe Billing is running significantly higher than usage of Stripe overall, because it skews toward people forming new software companies. That is not a claim about AI specifically. It is a signal that company formation at the small end is accelerating, and that the tooling available to tiny teams is finding users.

Quinn Slack, whose own company has stayed deliberately small, watches competitors his size begin hiring product managers and marketers and reads it as a category error. His view is that historically, 90 percent of a company’s headcount ended up in overhead: selling, coordinating, managing the distance between the people with product ideas and the people using them. Agents, in his framing, compress or eliminate that overhead layer, leaving a structure where the people thinking about the product are also the people building it.

Anton Osika adds a less institutional data point. He observes that employees at large companies are already using tools like Lovable, a development platform, to build side projects that generate hundreds of thousands of dollars before those employees become founders at all. The founding moment, in this pattern, is not the beginning of the experiment. It is the confirmation that the experiment worked.

Stefano Ermon, who leads Inception, a company of roughly 50 people, offers the plainest version of the shift. The bottleneck at his company, he says, is compute rather than people. That single inversion, from headcount to compute as the binding constraint, is what makes the Minecraft precedent feel less like a fluke and more like a template. Mojang’s extreme ratio of value to headcount was, at the time, treated as an artifact of the gaming industry and the particular cultural moment of Minecraft’s rise. The current evidence suggests the structural conditions that made it possible are becoming general rather than exceptional. The question for small teams is no longer whether the model can work. It is whether they will organize themselves to take advantage of it before the window narrows again.