How To Pick A Startup Idea

John, a YC partner, tackles a common founder paralysis: the inability to commit to a single startup idea. The core argument is that abstract deliberation without customer contact yields no useful data. He dismantles the 'perfect idea' fallacy, noting that true product-market fit is discovered through deep, unwavering commitment, not through a theoretical matching exercise. The most operationally relevant insight for technical leaders is the framework for going deep: 'burning the boats' on other ideas, rapidly becoming a domain expert capable of running a customer's business, and creating a tight build-measure-learn loop. For AI-focused teams, he provides specific evaluative criteria: the idea must sit at the edge of what models can currently do, it should 'verticalize' by owning an outcome (like an insurance product, not just software), and it should pursue the most ambitious version of itself because the effort cost is equivalent. The failure mode, pivoting, is reframed not as a setback but as a data-rich foundation for discovering the deeper, structural problem underneath the initial surface-level idea.