Is AI Breaking the Lean Startup Playbook?

Collison compares knowledge to a computer cache: recalling something yourself is faster than asking an agent and waiting for a response. He argues that cognitive ability still matters, and that students should not give up learning before there is evidence those benefits have been exhausted. He also says he still prefers to write his own messages because model-written prose has not yet felt compelling to him.

Collison dropped out of college twice to start companies, returning to MIT between ventures. He says students who enjoy college can finish, and that leaving need not be permanent; in his experience, people did not care about his dropout history. Looking back, he thinks his sense that startup opportunities might soon disappear was mistaken, since Silicon Valley has continued to offer opportunities over many decades.

Stripe began with a concrete frustration: online payments involved unpopular, outdated processes, even though banks initially found two young founders an unlikely fit for financial services. The company waited nearly two years for its public launch because payments required security, partners, reliable infrastructure, and money movement. It had a production customer within about two months, then built features such as a dashboard, refunds, and payouts in response to that customer's needs.

Collison questions whether the lean startup approach of finding a small niche and expanding from it is always right in the AI era. He suggests that those niches may be more crowded, while AI makes it easier to start organizations with broader capabilities. He points to AI labs and Anduril as examples of successful companies that took approaches he considers less lean.

Collison says large labs have limits because organizations struggle to pursue many priorities at once, though AI models may still replace some tasks or businesses. Stripe data showed new businesses starting on its platform at nearly twice the rate of the prior year, while median business performance and the chances of reaching revenue thresholds were also improving. He attributes faster startup sales in part to established companies fearing the cost of sticking with outdated ways of working, and sees signs of a more decentralized economy with many potential winners.