Garry Tan, CEO of Y Combinator, reports annual spending on AI model tokens between $50,000 and $200,000. He describes this strategy as tokenmaxxing, aiming to replace traditional engineering team output with AI agent workflows.
Tan released a toolkit called gstack on GitHub, containing 23 Claude Code skills designed to simulate an engineering team. The repository has gained between 100,000 and 136,000 stars from the developer community.
As a practical example, Tan utilized Claude Code agents to develop the Garry’s List blogging platform in five days. The project incurred a total cost of approximately $200 in tokens.
By treating significant AI token expenditure as a core operating cost, the Y Combinator head suggests a shift in startup economics. Founders may increasingly allocate budgets toward AI usage rather than traditional early-stage hiring.
