Garry Tan, chief executive of the startup accelerator Y Combinator, has put a number on his agents for the first time. "I spend about maybe $100[,000]-$200,000 a year on tokens," he said on an episode of Walter Isaacson's podcast The Builders, due to publish on Tuesday.

Tan is not a quiet user. He has joked in the past about suffering "cyber psychosis" and about barely sleeping because he enjoys working with his agents so much. He has published his 23 Claude Code skills as "gstack" on GitHub so others can learn what agents can do.

**What the agent reads.** By Tan's own account, not just project files. "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. The agent researches podcasts, takes on coding work — and helps run his life. Tan describes it as "a Jiminy Cricket on your shoulder", one that runs ahead, asks questions and sees around corners.

That is the actual finding of the week: the most extreme form of agent use is not the enterprise deployment, it is the personal one, and it is being lived by a man who hands out money to founders.

**And at YC itself.** The accelerator runs an internal agent called Stray Light, which Tan says "kind of has a personality" and has a "company brain" — the total context of every founder YC has ever backed, anyone who has ever emailed the firm, and all applications. The agent is "rebuilding the way we operate at YC," he said.

You do not have to judge that to notice what it means: data pooling that founders never imagined sharing with a chatbot — applications, rejections, confidential messages.

**The lesson he passes on.** Do not treat agents like traditional software, Tan said. People so worried about "repeatability and safety" end up trading away "all of the beautiful parts of working with an AI agent, which basically can be so similar to a human."

**And the cost problem.** Personal agents go mainstream, in his view, when one thing is fixed: "It's too expensive right now." He knows his own budget is "a crazy privilege that not everyone can do." Free agents run by the big platforms are not the answer either: "This is a huge company that controls the algorithm, and we all wish that they didn't at some level." The alternative, he argues, is a personal agent you control and program yourself.

Between the lines sits a business model nobody has built yet. An agent that has read half of a person's life is simultaneously the most useful and the most delicate product anyone could sell. Whoever operates it carries the liability for it.