President and CEO of Y Combinator and founder of Initialized Capital; early investor in Coinbase and Instacart, and co-founder of Posterous.
Tan's most jaw-dropping return: roughly under $20,000 of YC's early Airbnb investment became about $2 billion of liquid stock at the IPO, and YC didn't even take its pro rata.
“I mean, I think the craziest thing I'd ever heard was basically the Airbnb round where I think something like under $20,000 became $2 billion in the liquid IPO when Airbnb IPO'd. So, and you know, YC didn't even take the pro rata then.”
Tan previews unreleased YC data: investors who backed at least 3 companies per batch from 2018-2020 saw a median 5x, top decile 16x, and even the bottom quartile returned 3.3x, versus the typical VC top-quartile of only 3x.
“The wild thing though is median was 5x and then the bottom quartile was 3.3x. So it's really crazy because in venture period, if you look at how VCs make money, it's all in the power law. There's only, you know, sort of the top 25% make any money at all. And then the top quartile generally for any given year of VC, uh, is only 3x.”
Tan floats Chamath's 'capital as a service' idea: a founder connects Stripe, bank, and QuickBooks, an LLM runs the diligence and ranks whether the business is real, and if you clear a bar, money lands in your account automatically.
“like the capital as a service idea, like, is really just, you know, connect your metrics, connect your Stripe account, your bank account, your QuickBooks. You know, I mean, it's actually pretty doable today. Like, it's conceivable that someone should connect all those things up and then an LLM can just, you know, rank whether something's real or not, do the diligence. And then if you like hit a certain bar, like, boom, like, here's money in your bank account.”
Steal thisBuild automated underwriting: connect a founder's Stripe/bank/QuickBooks, let an LLM run diligence, and auto-fund anyone who clears a bar.
At 14, Tan cold-called businesses with internet listings in the Yellow Pages offering web design, landed a job biking to a firm that built city websites, and learned HTML and graphic design for $7-10/hour.
“So I just started cold calling, you know, seeing, hey, like, I know how to make web pages, will anyone hire me? And sure enough, I ended up getting a job and riding my bike to the local web design firm that made city websites for like City of Pleasanton and Fremont and San Jose.”
Tan describes the Peter Thiel / Palantir founding move: take an 8.5x11 sheet, write down every smart person you know, then take them to dinner and effectively sit on their doorstep until they quit their jobs to join you.
“It's like when you start a thing, you, you know, take an 8.5 by 11 sheet of paper and you write down all the smartest people, you know, and then you go take them out to dinner and lunch and just, you know, sit on their doorstep until they quit their jobs to come join you.”
Steal thisList every smart person you know on one sheet, then relentlessly court them over meals until they join your founding team.
Tan's biggest lesson from passing on Palantir: the world isn't narrated to you. Prime movers go out, talk to real users, and gather first-source information rather than taking their worldview from the Wall Street Journal or social media.
“Like, the world doesn't like happen to you if you are actually a like prime mover in the world, like Peter Thiel or my friends. Um, you go out and you go into the world and you discover something about the world. Like, you actually talk to people, like you actually talk to your users, and, uh, you I sort of gather primary first source information on how the world works.”
Steal thisDon't take your worldview from the press; go talk to real users and gather primary first-source information yourself.
Tan explains Palantir's origin insight: PayPal caught fraud rings by graphing identities (invisible to SQL/row-based methods), then realized the CIA and NSA lacked that graph technology entirely.
“And the core idea was that, um, they could find these fraud rings just using a graph database. Just literally, you know, people make a bunch of fake identities, they send the money from one fake identity to another, and, uh, if you're just using SQL or a row-based method, you just couldn't figure it out. But the second you like graphed it out, you would find this ring.”
Tan's blog platform Posterous grew 10x year-over-year twice until Instagram appeared; his co-founder dismissed Instagram as 'it sucks.' The real competition was how people got photos off new iPhones, not Tumblr, and they got run over by the eventual winner.
“And before I could, uh, reply, my co-founder replied, it sucks. I was like, and I was like, and we got a one-line reply from Sacca saying, I'm very disappointed in you guys.”
Tan contrasts Posterous with Weebly founders David Rusenko and Dan Veltri, who never raised a Series A, stayed profitable and cash-flowing, focused on users, and sold to Square for ~$300-400M, making roughly 100x more than Tan and his co-founder.
“So these guys are like made, you know, 100 times more money than me and my co-founder Impostor is like, you know, rightly so. They did it not through like continuing to raise venture dollars, but by, um, being profitable, cash flowing, by compounding, by focusing on their users and like playing their own game”
Steal thisPlay your own game: stay profitable, focus on users, and compound instead of raising rounds to look like a successful startup.
Tan argues that because Google and Meta hold a monopoly on eyeballs via second-price ad auctions, any non-creator must pay the toll for attention, so building your own audience is the only durable edge.
“this is sort of the reason why I feel like if you're running a business today, you kind of have to be a creator because if you're not a creator, you gotta buy eyeballs somehow. And, uh, guess who's got the monopoly on eyeballs? You know, Larry and Sergey and Mark Zuckerberg. You gotta go pay, you, you gotta pay the toll to, you know, Go across the bridge”
Steal thisBuild your own audience as a creator so you don't have to pay Google and Meta the toll for every customer.
Tan predicts that LLM-powered 'two-pizza teams' of fewer than 20 people will build $100M to $1B-a-year businesses, and that incumbent corporate America is completely unprepared to compete with these agile startups.
“Like, I think that we, you know, probably in your community, or maybe you guys yourselves will end up making these like $100 million to like billion-dollar-a-year businesses that are totally empowered by large language models that, like, do not need more than 20 people working at them.”
Tan predicts that LLM-powered 'two-pizza teams' of fewer than 20 people will build $100M to $1B-a-year businesses, and that incumbent corporate America is completely unprepared to compete with these agile startups.
“Like, I think that we, you know, probably in your community, or maybe you guys yourselves will end up making these like $100 million to like billion-dollar-a-year businesses that are totally empowered by large language models that, like, do not need more than 20 people working at them.”
Tan's debugging rule for AI apps: if the model hallucinates or can't reliably do the task, you're asking too much. LLMs do roughly 120-IQ work, so break the job into smaller steps rather than overloading the context window.
“and if there's hallucination or, uh, the LLMs aren't actually able to like consistently do the thing you want, you actually have to chop down the prompts more. Like you're asking for the LLM to do too much, like The LLMs are capable of maybe like 120 IQ level work right now. And if you're, you know, giving it like too much in the context window and asking it to output too much, it's just like, uh, do it in steps, right?”
Steal thisIf your LLM hallucinates, you're asking too much: chop the prompt into smaller steps instead of overloading the context window.
Tan's Burning Man allegory for bias to action: you must first believe you can change the world with your ideas, but belief alone is not enough. The people who 'bent the spoon' also used their hands. Go talk to users and build the thing.
“Like, real life is sort of like this exercise. Like, they say, like, you can go change the world and you can do it with your, you know, with your very ideas, but like You know, that alone is not enough. Like you actually have to go with your hands and go talk to users and, you know, sit and build a thing. And so, you know, don't forget the second half.”
Steal thisBelieve you can change the world with your ideas, then actually use your hands: talk to users and build the thing.