Summary
Stripe CEO Patrick Collison sits down with Harj Taggar at Y Combinator’s Startup School 2026 for a wide-ranging conversation that uses Stripe’s origin story as a lens on the AI era. Collison opens by wrestling with a question that will resonate with any developer: if a 16-year-old can prompt Claude to write a Lisp dialect today, should they still bother? His answer draws on Jeff Dean’s famous ‘numbers every programmer should know’ framework — arguing that cognitive L1 cache (internalized knowledge) still enables far faster reasoning loops than any AI tool, and that revealed preference at companies like Stripe and the frontier labs shows an ongoing premium on raw cognitive ability.
The conversation moves through Collison’s personal discipline of never using AI-suggested writing — he remains skeptical that language models have cracked interpersonal communication and essay-quality reasoning, despite their performance on formal proofs. On Stripe’s early strategy, he recounts the decision to delay public launch for nearly two years (first production user in January 2010, public launch September 2011) while operating in near-stealth, justifying it by the infrastructure, security, and banking-partnership requirements unique to payments.
For founders and engineers navigating the AI transition, Collison’s framing of knowledge as cache hierarchy offers a durable mental model for deciding what to learn deeply versus delegate — and his skepticism of AI writing quality provides a counterpoint to the dominant narrative about LLMs replacing knowledge work wholesale.
📺 Source: Y Combinator · Published July 31, 2026
🏷️ Format: Interview
