Y Combinator’s Garry Tan uses AI agents so much, he spends up to $200K a year on tokens

What changed

Y Combinator’s Garry Tan revealed he spends up to $200,000 annually on AI tokens to power personal AI agents. He shared this detail on Walter Isaacson’s podcast, The Builders, noting how intensely he interacts with these agents to the point of joking about “cyber psychosis” and losing sleep. Tan is not just experimenting; he has heavily integrated AI into his daily workflow, treating it as a core extension of his productivity.

Why builders should care

Tan’s high spending and deep reliance on AI agents underline the growing operational costs and intensity of using AI-driven workflows at scale, even for someone with his resources. For builders and operators, this signals that AI integrations are moving beyond simple augmentations and becoming substantial parts of routine work, with non-trivial token expenses attached. Managing token budgets and maximizing agent effectiveness has become a crucial operational concern.

The practical takeaway

Operators should anticipate significant variable costs when deploying complex AI agents and models persistently. This highlights the need for careful orchestration of token use and performance monitoring to balance AI utility against expense. Relying heavily on AI output requires infrastructure and budget planning not only for model access but for the fluid, continuous interaction AI agents demand.

What to watch next

Expect more transparency from leading builders on the real costs of deploying AI agents at scale, including token budgets and operational trade-offs. Watch for tools and platforms optimizing agent token efficiency and pricing models evolving to accommodate heavy continuous use. How startups and organizations measure AI-driven productivity against these growing costs will be key to wider adoption decisions.

AI Quick Briefs Editorial Desk