Y Combinator CEO Garry Tan has revealed that he spends up to $200,000 a year on AI tokens, using agents for everything from coding to personal life management. His candid admission highlights the growing cost and power of AI agents in high‑level startup ecosystems.
AI Agent Spending and Usage
- Annual token spend ranges from $100,000 to $200,000.
- Tan’s 23 Claude Code skills, dubbed gstack, are publicly available on GitHub.
- Agents ingest emails, Dropbox files, Apple Notes, and even 20‑year‑old therapy notes.
- He describes agents as a “Jiminy Cricket” that anticipates needs and offers proactive suggestions.
- Tan cautions against treating agents like traditional software, stressing safety and repeatability concerns.
Structured Overview
| Aspect | Details |
|---|---|
| Entity | Garry Tan / Y Combinator |
| Token Spend | $100k–$200k/year |
| Agent Name | Stray Light (YC internal) |
| Use Cases | Research, coding, life management, founder support |
| GitHub Repo | gstack (23 Claude Code skills) |
| Cost Implication | High; privilege for early adopters |
Future Outlook
Tan envisions personal AI agents becoming mainstream once costs decline. He argues that free agents from big tech may not suffice, advocating for user‑controlled, programmable AI. The internal YC agent, Stray Light, already aggregates data from every founder and application, illustrating how AI can reshape accelerator operations. As AI adoption accelerates, the industry will need to balance affordability, privacy, and autonomy to unlock the full potential of agents in both enterprise and individual contexts.