Often founders know what their problems are, but not their relative importance. They'll come in to talk about three problems — one of moderate importance, one that doesn't matter at all, and one that will kill the company if it isn't addressed immediately. It's like a horror movie where the heroine is upset her boyfriend cheated and only mildly curious about the door that's mysteriously ajar. You want to say: never mind about your boyfriend, think about that door!

Founders often misdiagnose their own problems. They'll come in worried about fundraising when the real issue is the company is doing badly and investors can tell. Or they'll worry about user acquisition when the product isn't good enough — sometimes admitting they wouldn't use it themselves if they hadn't built it.

Founders don't listen, and it's not just stubbornness — it's that so much about startups is counterintuitive. When you tell someone something counterintuitive, it sounds wrong, so they don't believe you until experience teaches them otherwise. A year later they come back and say 'We wish we'd listened to you.'

If you correct course at a high enough frequency, you can be simultaneously decisive at a micro scale and tentative at a macro scale. The result is a winding path executed very rapidly — like the path a running back takes downfield. A small improvement in navigational ability has a double effect: the path is shorter, and you can travel faster along it.

Speed defines startups. Focus enables speed. YC improves focus. Early-stage founders have a hundred problems and no one to work on them but themselves — so if they focus on what doesn't matter, nothing that matters gets done.

The educational system trains ambitious people to win by hacking the test instead of actually doing what it's supposed to measure. That stops working when you start a startup, so part of YC's job is retraining founders to stop hacking the test — and it takes surprisingly long; a year in, they still revert.

The 'batch that broke YC' in summer 2012 was an O(n²) algorithm problem: every partner had to know every startup, which worked at 60 but shattered at 80. The fix was sharding the batch into smaller groups with dedicated partners — proof that advising startups can't be pooled or automated.

Before YC, most assumed loneliness was the price of independence for founders. But great work historically clusters — Florence in the 1400s, Göttingen, Bell Labs, Xerox PARC — and very ambitious people need colleagues more than anyone, because they're so starved for them in everyday life.