How To Pick A Startup Idea

John, a partner at YC, opens by identifying a common founder failure mode: inability to commit to a single startup idea. He argues that meaningful progress is impossible without full commitment, and frames the video as a rubric for picking, committing to, and rapidly validating an idea.

He identifies two primary 'overthinks' that paralyze founders. The first is the search for a perfect idea, which he says cannot be found in the abstract — only through customer contact and real-world feedback. The second is excessive concern over founder-market fit. While he acknowledges that domain relevance matters, he argues founders routinely weaponize this against themselves, citing Blake Scholl of Boom Supersonic as an example of someone who successfully crossed domain boundaries from adtech to supersonic aviation.

John then addresses the trap of working on multiple ideas simultaneously. He argues this produces bad data: without going deep on any single idea, founders can neither confidently kill a good idea nor correctly abandon a bad one. His solution is 'burning the other boats' — explicitly shutting down all other options, changing company names and emails if necessary, and pursuing one idea with singular focus. He cites GovDash, which pivoted five times while fully committing to each iteration, as a model of this approach.

To measure whether a founder has gone deep enough, John uses the benchmark of whether they could run the customer's business themselves. Using voice customer service agents for cleaning companies as an example, he asks whether the founder knows daily operational crises, whether answering the phone is a top-five problem, and what revenue is lost per unanswered call. He also frames this as being able to teach a class on the problem or being among the most informed people in the world on the subject.

For validating ideas in the AI era, John outlines three qualities to look for. First, the idea should sit at the edge of what AI models can currently do — barely working today but clearly improving with model progress, referencing Paul Graham's 'live in the future' framing. Second, the idea should verticalize — meaning it should own outcomes rather than just sell software. He argues that as software costs approach zero, the valuable assets become customer trust, licenses, regulatory permissions, and outcome ownership. He cites Corgi Insurance, which acquired an insurance carrier during its YC batch rather than settling for being a tech-enabled broker. Third, the idea should be the most ambitious version of itself, since the cost of pursuing an ambitious idea and a modest one are roughly equivalent in difficulty, but the ambitious version offers better moats, talent attraction, and competitive protection.

Finally, John reframes failure: a deep dive that doesn't work still yields unambiguous customer data, real conviction for pivoting, and — most importantly — often surfaces a better, deeper structural idea that wasn't visible at the outset. He concludes that the worst failure mode isn't being wrong, but spinning wheels across multiple ideas without going deep enough on any of them to learn.