Y Combinator’s Garry Tan uses AI agents so much, he spends up to $200K a year on tokens
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
Running high-context, autonomous agents that ingest complete organizational histories and multi-decade personal data can cost a single power user up to $200,000 annually in LLM tokens. For engineers deploying enterprise agents, this extreme cost curve means that building "company brain" systems with continuous, un-truncated context is financially non-viable at scale without aggressive context caching, semantic retrieval routing, or hybrid local-cloud architectures.
A power user running AI agents at near-maximum intensity is burning $100K–$200K/year on tokens for both coding and full life-automation (email, Dropbox, 20 years of notes fed as context), which signals that today's frontier-agent economics still don't pencil out for mass or even heavy individual use. The practical takeaway: per-token cost remains the binding constraint on autonomous-agent deployment, so budget aggressively and treat broad context-loading as a real line item—and note the explicit advice to stop over-engineering for repeatability/safety, since that defeats the probabilistic, human-like behavior that makes agents useful.