Garry Tan’s $200K AI Agent Habit: Cost, Access, and Privacy

Y Combinator CEO Garry Tan says he spends roughly $100,000 to $200,000 a year on AI tokens for agents that help with coding and work. He also says his personal agents have read years of email and notes. His estimate is a self-report, and no data breach was reported.

This content is blocked because it would connect to YouTube.

How much does Garry Tan say he spends?

Tan estimated that he spends about $100,000 to $200,000 annually on AI tokens, according to Julie Bort’s TechCrunch report. The figure is his rough estimate. It is not an audited expense, a typical user’s bill, or Y Combinator’s total AI spending. He says agents help him with coding and many work projects, and he has shared 23 Claude Code skills under the name gstack.

What can his personal agents access?

Tan says he has given personal agents access to his email, Dropbox, Apple Notes and even therapy notes reaching back two decades. That is his description of permissions he chose to grant. TechCrunch did not report a breach or a leak of those records. The practical question is how much private context someone should share for a more useful assistant, and what controls they have over that access.

How is Y Combinator’s Stray Light different?

Tan separately described an internal Y Combinator agent called Stray Light. He says it can draw on company context such as founder histories, contacts and applications to help prepare for interviews and suggest next steps. Stray Light is distinct from the agents Tan uses for his personal work. The report does not say it makes investment decisions on its own or that company records were exposed.

Could personal agents become affordable?

Tan calls his own spending an unusual privilege and argues that cheaper, user-controlled personal agents could become common. That is a forecast rather than a measured outcome. The tradeoffs to watch are token cost, which data an agent can read, how to revoke access, and whether the service gives users meaningful control. Watch our Short or listen to the full episode. This discussion is based on TechCrunch’s November 2 report; the Short’s graphics are illustrative.

Leave a Comment

Your email address will not be published. Required fields are marked *

Save to list

Scroll to Top