Key points
- Y Combinator CEO Garry Tan says he spends roughly $100,000 to $200,000 a year on AI tokens, per a Walter Isaacson podcast episode.
- Tan has shared gstack, a set of 23 Claude Code skills, on GitHub to help others learn what agents can do.
- YC runs an internal agent called Stray Light that has context on every founder it has backed and all applications.
- Tan says personal agents will go mainstream only if costs fall; he calls his own spending a privilege.
Why it matters for marketers: A top startup leader puts a number on heavy agent use and says cost is the barrier, a useful benchmark for teams budgeting agent workloads.
Y Combinator CEO Garry Tan says he spends about $100,000 to $200,000 a year on tokens to power his AI agents. He gave the figure on Walter Isaacson's podcast The Builders, in an episode due out Tuesday, according to TechCrunch.
Tan uses agents for work projects, coding and daily life. He says his agent has read his email, Dropbox files and Apple Notes. He has also published gstack, a set of 23 Claude Code skills, on GitHub to show others what agents can do. Inside YC, an internal agent named Stray Light holds context on every founder the accelerator has backed, everyone who has emailed it and all applications.
For marketers and content teams, the number is a rough signal of what an intensive, always-on agent setup can cost, and of how far one executive's usage sits from typical budgets. Tan himself says the spend is a privilege and that personal agents will not go mainstream until costs come down. He also argues teams should not treat agents like traditional software, and that over-focusing on repeatability and safety can sacrifice much of what makes agents useful.
Worth watching: whether token prices fall enough to make company-wide agents practical for smaller teams, and whether the idea of a personal agent that users control and program, which Tan raised as an alternative to Big Tech algorithms, gains traction. Teams experimenting now may want to track token spend per workflow against the output it delivers.
Sources
This brief is our own summary. Read the full reporting at the source.
