Garry Tan

Garry Tan fears a future in which a light cone of commercial value collapses into five AI kings. Friends inside some of the AI labs, he says, were already describing that picture last year: anything a business might do could end up inside five companies. Walter Isaacson had asked whether Anthropic, OpenAI, and Gemini would consolidate to two or three. Tan says such a consolidation would be pretty horrible, and that the field is already there.

He wants personal AI instead, closer to the personal computer than to a social feed. A feed run by a big tech company has a business model that works best when it enrages or divides people, and the company keeps the control. Personal AI, as he hopes it

Garry Tan says Y Combinator vets a founder in a 10-minute interview that ends in an up-or-down decision. The firm now gets 80, maybe 90,000 applications a year for about 800 spots, an acceptance rate under 1 percent. Maybe 8 to 10 percent of applicants get an interview, sometimes less. Paul Graham figured out that a half hour or an hour was unnecessary. Five minutes would probably do, and five minutes feels rude, so the interview stayed at 10 minutes.

The limit works like the old 140-character cap on Twitter. In the window, Tan wants a real person, a problem, and a plain account of how the technology solves it, plus why the idea is worth building. If a little time is left, he wants to know

Garry Tan wants more billionaires. In his view, saying billionaires should not exist is also saying society should ossify, an LLM should not exist, and Google should not exist.

He does not begrudge Google, even though some people on the hard left say Google should not exist. He wants more companies like Google. OpenAI was a spin-out of Y Combinator. Sam Altman sat in the same seat, saw the front row of young people with crazy ideas, and with Greg Brockman created YC Research that became OpenAI. Life now runs on an accelerated timeline, the age of intelligence is much closer, and he does not think that is a bad thing. AOC and a great many Americans believe all billionaires are ill-begotten.

Garry Tan says the responsibility for CEOs and people who have a lot of privilege is not to cost cut. Cost cutting is the easy way, and it can result in a dystopian moment where capital accumulates more and more wealth and people are not sharing it.

Someone with domain expertise should pair with a 22 year old who knows how to make 400 times the amount of software people could make before, or train themselves to do it, and start a company in the space they know. The company may be exactly what they did before, or something two clicks away. Either way, it requires an increase in the amount of agency. People should go into their own history and their own lives and ask what they uniquely know,

Garry Tan says an AI agent's working memory can already hold about a thousand pages, roughly three Harry Potter books, and pull any page, fact, or event from that stack. A person can keep track of only seven plus or minus three things at once. The mistake he names is treating AI as chat. An agent is far larger than a chat window, and Y Combinator is starting to rebuild the way the firm operates around one.

Straylight, an internal AI agent, has a personality and a company brain: the total context of every founder Y Combinator has backed, anyone who has ever emailed the firm, and all the applications. He walked through the scale with his 11-year-old. The agent can find a needle in that haysta

Garry Tan says groups that claim to speak for little tech are paid shills, funded with millions of dollars in astroturf money.

Most tech people stay out of politics, but politics stays interested in them, so three or four years ago he started spending time on it. The groups used almost hired actors who went to Sacramento and D.C. and said the little guy wanted rules that mainly help big tech. Real startups were not talking to people in D.C. He saw a broken system and wanted to fix it. Y Combinator is not a union, he says, but it has union-like characteristics, and it is taking a seat on the billionaire tax in California and on antitrust laws in D.C. and Europe so the golden goose can still

Garry Tan says true free market enterprise has never been tried. Google was able to exist, thrive, and become a trillion-dollar company because of the DOJ consent decree on Microsoft.

Microsoft was ready to take over the web and used its platform advantages in Windows to default install a browser and take over the browser market. Google is now doing the same thing. Delaware C Corp governance tells leaders their duty is to maximize shareholder value, so people follow the letter and the spirit of that structure and close platforms. On a Google search, less and less of the screen is even non-Google properties. The pattern is self-preferencing.

Self-preferencing should stop, and markets should

Garry Tan learned from Alex Karp that the hotel tells a buyer what Palantir is selling. If the team says Motel 6, the buyer thinks they are buying Motel 6 software. Palantir is Mandarin Oriental software.

At 25, he was sent to DC in ripped jeans and a band t-shirt to demo for customers and was booked at the Mandarin Oriental in D.C. He and the other early engineers on the Palantir Finance product wondered why they were staying at the Mandarin Oriental. They were just kids at a startup and asked to be set up at the Motel 6. They were going to talk to a general or a VP of some Fortune 500 company, show the software they had built, then make small talk about how long they were around and where

Bookface is internal software at Y Combinator. On The Peel, Turner Novak had heard about Bookface and asked whether Garry Tan made it. Garry Tan said yes. Y Combinator started becoming a thing, people started pretending to be Y Combinator companies, and there was no way to verify them.

Garry Tan says Y Combinator had a Google group or something. Vibe coding was not a thing back then, so Garry Tan real-coded Bookface in Rails. At the start, Bookface was actually just a Facebook. The name is a little bit of an inside joke from a skit on The Office. Garry Tan says the joke is of a certain generation, and Garry Tan likes it. The 22-year-olds do not watch any of the stuff that generation came up

The weekly growth chart on bathroom Post-it notes is why Airbnb raised from Sequoia before demo day. Garry Tan says they needed revenue on the site to grow 10% every single week, and they nailed it.

He says the number one thing is that objects in motion stay in motion, and objects at rest stay at rest. Y Combinator is the moment founders learn to run fast. Billion dollar companies look back when a thousand person company is going slow and try to tap the energy of the first week and the first eight weeks. Airbnb got rejected by all the legends too. Brian Chesky used to obscure who the people were. Mike Maples later posted that he was the one who said no. He is one of the biggest legends in v

Garry Tan says Y Combinator just extended the program to 13 weeks. The team talks to people about how things went and what can be done better. Moving from two batches to four made each batch a little shorter, down to maybe 11 or 12 weeks. People needed a little more time, so Y Combinator added about two more weeks and set the program at 13.

The first nine to ten weeks are for finding product-market fit. The work is to get into the market, nail the messaging, figure out who the product is for, build it, and land a set of real customers. Sometimes the buyer is an enterprise. The first three to five customers matter a lot, and early sales set the tone for the whole company.

Developer tools be

Two or three years of YC time as a partner is equivalent to 20 years in venture. The 15 partners are practitioners of product market fit. Tom Blomfield created Monzo, worth billions of dollars and one of the top fintech companies in the world. People on the team have sold companies for hundreds of millions of dollars.

The reps are why the organization learns. A partner reads a thousand to two thousand applications every single year and writes 50 seed checks in a given year, sometimes 75. Pulling out the billion dollar idea for founders is hit or miss and requires inspiration. Paul Graham and Vinod Khosla are the standard, and all of the YC partners are in the top two to five percent in the

On The Peel, Y Combinator gets about 80,000 applications a year. Something like 78,500 of them are mortal wounds, the rejections handed out. Between 700 and 800 companies get funded in a year, about 1%.

Asked whether 80,000 is per year, the answer is yes. A guess of 200, 250, or 300 companies per batch is met with the yearly total, between 700 and 800. The partner does not like handing the nos out, and the follow-up is please apply again. An Adventure Time meme the partner tries to post says getting rejected from Y Combinator is the first step to being accepted to Y Combinator.

Handing out the rejections keeps the partner up at night. The meme is the other half of the same point: apply aga

Garry Tan says 99% of people who apply to Y Combinator end up getting a mortal ego wound by being rejected. The rejections keep him up at night because the wound is not a good thing. Something like a third of the batch, and sometimes half of the batch, will have been rejected before getting in. The number has been going up. People quite commonly apply two to four times before they get in.

Y Combinator is always worried about over-rotating. Being too small can be bad, and being too big can be bad. Garry Tan says there are way more really capable founders being rejected right now. Y Combinator should not ask people to be rejected time after time. A founder who survives the mortal ego wound on

Garry Tan says Y Combinator is a managed marketplace for the smartest people starting the next companies, and he calls the setup a good thing. Y Combinator is about 20% of all companies worth $5 billion and up that started from 2012 and later. The share has stayed about 20% for 15 to 18 years because Y Combinator is a network effects business.

The marketplace view came from a concerted effort over the last two or three years. Garry Tan says most of his VC friends would agree. He wants a bigger share. Mark 2026 on the calendar: if he is successful in five or 10 years, he would like the share to be 30%, and he would like the share to be 50%. He asks why the share cannot be 30 or 40. The aim i

Initialized Fund One was a $7 million fund that returned more than 55x DPI. Most of that return came from Coinbase and Instacart.

The fund backed a hundred companies. The DPI on 98 of them added up to about 1x. Instacart was another 1x, and the fund could not take any pro rata because Sequoia did. On Coinbase, the fund took the pro rata in multiple rounds and ended up turning the fund 53x.

The power law is astonishing and unbelievable. Reading about a return like that and hearing about it still leaves it hard to believe until one company does nearly all the work.

Garry Tan joined YC with no plan to invest. After he raised and quit his startup, he was really burnt out and needed to recuperate. He had no intention of ever becoming an investor, ever becoming a GP, or starting a VC fund.

It was still the era of consumer startups, when people said a co-founding team needed a designer. He was told he was a great designer and asked to help a bunch of YC companies design their homepages. He did so many office hours that his pet theory is maybe they saw how much work he was doing. They were already promoting other great people into investing partner, and they said they kind of needed to promote him too. He says he snuck right in. Asked whether he was actuall

On The Peel, Dropbox becoming a billion-dollar startup around 2011 was called one lucky year. Airbnb becoming a billion-dollar company in 2010 and 2011 was the second hit, and YC then went from giving founders $12,000 and $15,000 to $100,000, $120,000, and $175,000.

Garry Tan was a software engineer at Palantir, employee number 10, and he designed the Palantir logo. Palantir had only 10 people and tried to hire YC alumni, which is how he found out about YC. YC started in 2006 and 2007. Back in 2007 and 2008 he already saw YC as where smart engineers went to start a company, and he wanted to be like Steve Huffman, Drew Houston, and James Lindenbaum, who created Heroku. Around 2011 Dropbox ha

Garry Tan sent an understanding of an actual proposition to a number of reporters, and the reporters would not publish it. The reporters did not think the proposition was noteworthy. The refusal astonished Garry Tan. Garry Tan says it does not feel like the media environment is designed to support the idea that tech should exist at all. Garry Tan thinks a positive role for tech is so unpopular with certain sets of people in society that the most mainstream business publications are starting to refuse to publish an argument against an asset seizure.

Garry Tan calls going direct on the internet the counterbalance. Garry Tan did not want to break the proposition on X. The conversation goes on

On the show, Garry Tan says the program started as an experiment by Paul Graham and Jessica Livingston to give small amounts of money, almost summer-internship style. His own batch paid only about $13,000. Twenty years in, YC is going on more than 100 billion-dollar companies.

People apply online, so a founder does not have to know anyone. YC accepts about 1% of the people who apply. Getting in brings half a million dollars for about seven percent of the company, and he says a YC partner matters more than the money. He is one of 15 people who select companies: they read the application, watch the video, meet the team for 10 minutes, and decide yes or no on half a million dollars within the