Latest Exclusive Interview with Garry Tan, Head of YC: I Missed Palantir in 2003, and 22 Years Later I Summed Up the New Rules for Founders

At the age of 22, he turned down Peter Thiel, because "I could probably level up to level 60 at Microsoft this year."

This decision cost him a multi-billion-dollar opportunity — early equity in Palantir. Today, as the president of YC (Y Combinator), Garry Tan regards this experience as his most valuable lesson and repeatedly shares it with the new generation of founders.

In a recent interview, the most powerful leader of Silicon Valley's incubator opened up about everything: from the job-hunting struggles after the dot-com bubble burst in 2003, to his scathing criticism of the "role-playing" culture, and the new survival rules for founders in the AI era.

I. 2003: The tech industry "died", and he went to Microsoft

In 2003, the Nasdaq had just crashed, and Silicon Valley was in a full slump.

"I literally couldn't get a job anywhere in the Bay Area," Tan recalled, "except for a position at Gap." His two offers both came from Seattle — one from Microsoft's Windows Mobile team, the other from Expedia.

He chose Microsoft, because he "believed mobile devices would be the next big trend". But it was a decision he still regrets to this day: at that point, he already had 5 years of web programming experience, and could have stayed in Silicon Valley to ride the upcoming wave of the social network boom.

"Mark Zuckerberg was building Facemash around 2003, and Facebook hadn't even been born. But I left web programming right at that critical moment," Tan said, "I did exactly the wrong thing at exactly the wrong time."

He named the lesson he learned for himself: Don't LARP.

"I was 22 years old, and I was swayed back and forth by what I read in the Wall Street Journal," he said. People always ask him "What's hot right now? What should I do?" — That's the wrong question. The right question is: "What do you have a unique interest in? What do you know more than other people?"

"If you have 5 more years of experience in a certain field than everyone else, you can completely outmaneuver everyone else."

II. A $700,000 mistake (actually billions of dollars)

An even more painful lesson was yet to come.

Tan's friends from his Stanford fraternity, Joe Lonsdale and Stephen Cohen, were interning at Peter Thiel's hedge fund. Thiel had a habit: when you start a company, write down the names of the smartest people you need to recruit on a piece of paper. They wrote down Tan's name.

Thiel invited him to dinner, handed him a check for $70,000 — equal to his full year's salary at Microsoft — and said: "I'm certain this is the right thing for you to do."

Tan said: "Thank you, Mr. Thiel, but I could probably level up to level 60 at Microsoft this year."

He did reach level 60. But Palantir was later valued at tens of billions of dollars.

"That was probably a $2 billion to $4 billion mistake," Tan said with a bitter smile, "and it was the exact same mistake as before — I made decisions based on the 'map' again, instead of observing the 'terrain'."

The so-called "observing the terrain" means accessing real, first-hand information. And "looking at the map" means listening to other people telling you what's hot and what's not. "Joe and Stephen are some of the smartest people I've ever met. If they weren't worth working with, who would be? But back then I only cared about what was cooler, what would make good investors say 'hmm'."

III. The essence of YC: turning outsiders into insiders

From the Palantir experience, Tan derived a core insight: "All the great things in my life are like some kind of 'cult' — they start with a truth or belief that goes against conventional wisdom, and they're extremely punk."

This is also the soul of YC.

"Paul Graham and Jessica Livingston found that there are a huge number of extremely smart people who cannot get access to Silicon Valley," Tan said. "In the old days, you had to go to the right schools, know the right people, attend the right events. What YC does is very simple: a website, 12 questions, and later a 1-minute video added. On the other end is a group of people who really care about your ideas and execution capabilities."

He calls it the "birthright of the tech industry" — every year YC's startup school attracts 7,000 people to San Francisco, most of whom are visiting the Bay Area for the first time.

"In Silicon Valley, where you come from and who you are really doesn't matter. What matters is your idea, and whether you can execute on it."

IV. New rules for founders: code is no longer scarce

Tan stated bluntly: In the past two years, the rules of the startup game have completely changed.

"Pure SaaS — the model that charges per seat — I'm not sure it will still exist 5 to 10 years from now. It's fine as a 'wedge', but if you're still building pure SaaS in 2026, you'd better pray it leads to some kind of data moat or network effect."

What's the core change? Code is no longer valuable.

"Back in the day, to get a product online, you had to go through all kinds of hardships — write product specs, find engineers, do QA. That made you feel like 'what I'm doing is important'. But today, you type /qa, and it's done. We have unit tests, 90% test coverage."

He described a brand new way of working: "One markdown file is an employee. And this employee will complete the task perfectly every single time, as many times as you want to run it."

V. Live in 2028: get superintelligence with "token maximization"

Tan shared his own practice: through open-source tools like OpenClaw, he makes his AI agent "almost reach an IQ of 150 on every single request".

"This requires you to allow yourself to 'maximize tokens'," he said. "It's completely reasonable for a founder or CEO to spend $50,000 to $100,000 a year on agents. Because it lets you live in 2028 today."

With this capability, a founder's efficiency can be amplified 400 times. "A 35-year-old or 45-year-old founder who has seen the world and built many companies, suddenly has 400 copies of 'himself' around him — he can outperform the entire department of any of the 'Big Seven' tech giants."

The key is: Every time you complete a complex task, turn it into a markdown file + code + test, and put it in a scheduled task. Next time it will run automatically. If there's an error, you just fix the bug once, and it will work correctly forever.

"Do every process your business needs perfectly once. The first time will definitely be bad, it will take a lot of iterations, but eventually you will have a digital employee that can be replicated infinitely."

VI. Conflict, management and the "7±2" rule of human nature

Tan believes Most businesses fail not because their product is bad, but because their management collapses — the scale of the business exceeds the capacity that any single person's brain can hold.

"Humans can only remember 7±2 things at any given time," he said. "But with agents, you can fit roughly three volumes of Harry Potter in your head."

He cited the example of Pedro, founder of Brex. Pedro built a layer of open-source tools (called Craptrack) with OpenClaw, letting agents monitor all network traffic and OpenClaw's behavior, so that he can safely use AI agents in a regulated fintech company.

More importantly: Pedro lets agents review the meeting notes of all his direct reports, so he can observe conflicts and problems two levels down in the organization. When he walks into a meeting, he already has the full context of what the compliance team has been working on for the past three weeks, and he can say on the spot: "You're right, do it your way." Then walk out.

"That's super powerful," Tan said. "Do you know that's a common reason for business failure? — being unable to track conflicts inside the organization. But now you can get 'X-ray vision' of your organization."

VII. "White Pill": social change is slower than technology, and that's a good thing

When asked about the anxiety brought by AI — white-collar unemployment, a permanent underclass — Tan's reaction was unexpectedly optimistic.

"The actual speed of change for all these things will be much slower than you think. Society is slower than you think, governments are slower than you think, every company in the world is slower than you think."

He calls it the "White Pill" — not the Pink Pill (false optimism), nor the Black Pill (total despair), but recognizing that things are complex but still choosing to take action.

"Most people in Silicon Valley say 'I'm going to disrupt them'. The reality is that they won't. There are structural reasons why Microsoft won't disappear. And that's not necessarily a bad thing."

VIII. From algebra to responsibility

Tan also shared why he devoted himself to local San Francisco politics in recent years. During the pandemic, he saw two things happen at the same time: district attorneys turned a blind eye to violent crimes targeting elderly Asian people; school boards openly showed hostility towards Asian kids who wanted to learn algebra in public middle schools.

"That hit me hard," he said. "I was that kid who went to public schools in Fremont. If you can't learn algebra in middle school, you can't learn calculus in high school, then I wouldn't have been able to study engineering at Stanford, and I wouldn't be where I am today."

He and the community organized, went to farmers' markets, and protested at ABC News stations. "When the system becomes so broken, you have to physically stop it."

"San Francisco is getting better now," he said. "Things are much better, everyone can feel it."

Conclusion

Garry Tan's story is a cycle of mistakes, reflection and action. He missed Palantir at 22, became a YC partner at 34, and now he tells founders every day: Don't chase trends, don't LARP. Be honest about your own experience, observe the real terrain, then amplify yourself 400 times with AI.

His core belief is very simple: "We may never reach utopia, but the act of trying itself is worth it."

This article is from the WeChat public account "Postwave Evolution Planet", author: Mark, published with authorization from 36Kr.