Y Combinator’s Garry Tan uses AI agents so much, he spends up to $200K a year on tokens

Y Combinator CEO Garry Tan disclosed he spends roughly $100,000 to $200,000 annually on AI tokens, putting a price tag on his heavy use of agents during an appearance on Walter Isaacson's podcast The Builders, set to publish Tuesday. Tan has previously joked about suffering "cyber psychosis" and sometimes barely sleeping because he's so excited to work with his AI agents.

The spending spans both work and personal life. Tan uses agents for research — including prepping for questions on Isaacson's podcast — and for coding tasks, and he has shared his set of 23 Claude Code skills, called gstack, on GitHub to teach others what agents can do. He says one of his agents has ingested all of his email, his Dropbox, his Apple Notes, and even two decades of therapy notes, describing the experience as having "a Jiminy Cricket on your shoulder at all times" that runs ahead and helps him see around corners.

Tan, an early OpenClaw adopter and now a power user, argues that treating AI agents like traditional software misses the point. He said some people are so focused on "repeatability and safety that they actually trade off all of the beautiful parts of working with an AI agent, which basically can be so similar to a human." YC itself is leaning into that philosophy: the accelerator runs an internal agent called Stray Light, which Tan says has "a personality" and a "company brain" holding the total context of every founder it has backed, every person who has emailed it, and all applications — an effort he describes as rebuilding how YC operates.

Tan believes personal AI agents will go mainstream once one blocker clears: cost. "It's too expensive right now," he said, acknowledging that his spending level is "a crazy privilege that not everyone can do." He's also wary of free agents controlled by Big Tech, drawing a parallel to Meta's algorithmic control over social media: "you could imagine a personal AI that you control, that you program."

The open questions for builders are whether token costs fall fast enough to make Tan's usage pattern affordable at scale, and whether user-controlled personal agents can compete with subsidized Big Tech offerings — or whether the agents that win will be the ones that hold the deepest context, like Stray Light holds YC's.