Y Combinator CEO Garry Tan spends up to $200,000 a year on AI agent tokens

YC chief puts a price tag on living with agents

Garry Tan, the CEO of Y Combinator, estimates that he spends somewhere between $100,000 and $200,000 a year on tokens to run his AI agents. The figure emerged in a conversation with Walter Isaacson on the podcast The Builders, scheduled for release on Tuesday, as reported by TechCrunch.

Tan is an unusually heavy user even by Silicon Valley standards. TechCrunch notes he has joked in the past about suffering from "cyber psychosis" and getting little sleep because he is so absorbed in working with his agents. Beyond his own usage, he has published a collection of 23 Claude Code skills, called gstack, on GitHub, which he describes as an effort to teach others what agents are capable of.

The agent has read everything

Speaking to Isaacson, Tan said his agent is woven into both his work and his personal life. It helped him research the podcast appearance and rehearse the questions he might face, but it has also ingested an enormous amount of personal material. "It's read all of my email and all of my Dropbox. It's read, like, all of my Apple Notes, even my therapy notes going back 20 years," he said.

He framed the experience as having a constant companion that knows him well enough to anticipate what he needs — a "Jiminy Cricket on your shoulder" that proactively suggests options and, in his words, helps him "see around corners."

Tan, whom TechCrunch identifies as an early adopter and now a power user of OpenClaw, argued that agents should not be managed like conventional software. In his view, teams that insist on strict repeatability and safety guarantees give up the very qualities that make agents useful, because a good agent behaves in ways that are "so similar to a human."

An agent inside YC itself

Y Combinator is applying the same philosophy internally. Tan told Isaacson the accelerator operates an agent named Stray Light, which has a personality and what he called a "company brain": complete context on every founder YC has backed, everyone who has ever emailed the organisation, and all of its applications. He said the agent is "rebuilding the way we operate at YC."

Cost is the thing standing in the way

For all his enthusiasm, Tan does not believe personal agents are ready for everyone, and the reason is price. "It's too expensive right now," he said, adding that his own token bill is "a crazy privilege that not everyone can do."

Nor does he see free, big-tech-controlled agents as the answer. Discussing Meta's influence, he argued that people already resent a huge company having control of the algorithm, and sketched the alternative he would prefer to see built: "a personal AI that you control, that you program."

Why it matters

Tan's budget is one of the few concrete public data points on what sustained, agent-driven AI usage actually costs. If a single power user can run up a six-figure annual token bill, it signals both how much compute agentic workflows can consume and how far prices need to fall before they are within reach of average users. The episode also surfaces two questions the industry has yet to settle: how much personal data people will be willing to hand over to agents that read their email, files and notes, and whether the eventual mass-market agent will be a service controlled by a large platform or something the user owns and programs directly. As the head of Silicon Valley's most influential startup accelerator, Tan's working habits tend to foreshadow where the rest of the industry is headed.