Y Combinator cofounder Paul Graham spoke at the Launch conference in San Francisco yesterday afternoon in a “fireside chat” with the event’s founder, Jason Calacanis.

While it wasn’t exactly a hard-hitting interview – these things are rarely intended to be – Calacanis managed to surface a lot during their conversation. Graham spoke at length about about Y Combinator’s earliest days, for example; addressed a couple of the controversies he has found himself embroiled in over recent months; and explained the rationale behind his decision to relinquish day-to-day control over Y Combinator. (He’s basically exhausted and wants his “brain back.”) Calacanis also asked Graham plenty about what indicates to him that a startup might succeed or fail. Here’s some of what Graham — whose platform has helped launch Airbnb, Dropbox, and Stripe, among more than 600 other companies — had to say:

On one of the quickest ways to get crossed off the list during an application interview with Y Combinator:

“The founders have to get along. If the founders hate each other, you’re in big trouble,” and it happens “very, very often,” said Graham. “You don’t know how good friends you are with somebody until you try to start a startup with them. That’s why it works so badly when you have some startup that’s started by some dude in business school who has this idea for some startup, and then he goes and finds some, like, 20-year-old meek, undergraduate computer science major to realize his vision, and that’s the founding team … If you go into a Y Combinator interview, and one of you looks in terror to the other one before answering questions, that’s one of our secret tells. Or if you roll your eyes while your cofounder is speaking, which has actually happened, or if you stand up your cofounder – like you don’t show up for the interview… these have all happened.”

On a founder type that Graham may have misjudged earlier on his career:

“The one thing is people who are very smart, but that’s it. People who are very smart but ineffectual. We used to have more faith in brains. It turns out you can be surprisingly stupid if you’re sufficiently determined. And anyone can tell this empirically. There are some parts of America where there are a lot of rich people and they’re not very smart – parts of Manhattan and Florida and L.A. You don’t have to be supersmart if you’re fearsomely effective.”

Calacanis asked him the most important thing for startups to focus on:

“There’s a meta answer to that,” said Graham. “The most important thing for startups to do is to focus, because there are so many things you could be doing, but one of them is the most important, so you should be doing that and not any of the others. So you should not be grabbing coffee with investors. When you want to raise money, you shift into fundraising mode and you go and raise money. You do not promiscuously meet with investors in the middle of the day when you should be working simply because they send you an email saying, ‘Hey, let’s grab coffee.’ There are a 1,000 things you could be doing, and only one of them is the most important … and you work on that.”

On the essence of growing a startup:

“You have to start with a small, intense fire. Suppose you’re the Apple I. I think they made something like 500 of those things. So all they had to do was find 500 people to buy these things and they launched Apple. Apple! So you’ve got to find a small number of people – it’s necessarily going to be a small number of people…who want what you’re making a lot… You don’t have to do any better than Apple and Facebook. You’ve got to know who those first users are and how you’re going to get them, and then you just sit down and have a party with those first few users and you just focus entirely on them and you make them super, super happy.”

Calacanis also managed to back into a question about how Graham righted the ship when, in 2012, it began to seem that Y Combinator was accepting too many startups into the program for its own good. (Its summer 2012 class welcomed 80 teams.) Considering that Y Combinator intends to grow much bigger, and may even spread to other cities eventually, according to Graham, his answer seems noteworthy:

“We thought, ‘Why does this batch suck so bad? Why do we hate our life?’ There were startups, like halfway through the batch, I still didn’t know what they were doing. And we were asking what went wrong and it was so obvious…It was N squared, specifically,” said Graham. “It would be no problem having that many partners dealing with that many startups, so long as they were sharded,” he added, referring to a programming word that means a horizontal partition. “So we redesigned YC to be sharded and it has been ever since and it works just fine. Like, 68 [teams], no problem … We have three siloes, each one overseen by a group of partners. So basically, it’s like three little Y Combinators. And we know little Y Combinator works.”

Sign up for our morning missive, StrictlyVC, featuring all the venture-related news you need to start you day.