Summary
Garry Tan, President and CEO of Y Combinator, delivered a keynote at the AI Engineer conference arguing that AI-native organizations represent a structural reinvention of how companies work — not merely a productivity upgrade. Tan opened with his widely debated 400x personal productivity claim, methodically deflating objections himself: even assuming bloated code, generous self-scoring, and maximum penalty for verbosity, he argues the floor is 8x and the midpoint is around 40x.
Drawing on internal YC data, Tan revealed that in the Winter 2025 batch a quarter of portfolio companies had codebases that were 95% AI-generated — and that cohort has become the fastest-growing, most profitable in YC’s history, with 94 companies total having crossed $100 million in revenue from a seed check. His key insight: the differentiator is not which model founders use (everyone accesses the same Claude weights), but how they “wire the work.” He maps agent architecture directly onto org structure — skill files as employees, resolver tables as org charts, filing rules as internal process docs, and trigger evals as performance reviews.
Tan also introduces a practical framework distinguishing “latent space” computation (taste, judgment, and ambiguity resolution handled by LLMs) from “deterministic space” computation (structured code execution), arguing most AI engineering failures stem from putting computation in the wrong space. The talk closes with a concrete example of seating 800 startup school attendees optimally — a task requiring both spaces in tandem — as a model for the kind of compound problem-solving now within reach of a one-person team.
📺 Source: AI Engineer · Published July 17, 2026
🏷️ Format: Keynote Launch
