Y Combinator chief urges founders to splurge on AI token usage

On an episode of the A16z podcast, Garry Tan, chief executive of Y Combinator, advised founders to adopt a "burn, baby, burn" mindset toward AI token usage, even if it means spending up to $100,000 a year. He believes that loading massive token counts—such as a million or 800,000—lets companies experience capabilities that may only become mainstream by 2028. Tan recommends first letting agents consume the resources needed to complete tasks, then documenting the process in markdown files that act as repeatable employees.

This stance opposes the criticism of "tokenmaxxing" voiced by Uber’s technology leader Praveen Neppalli Naga, who calls for more efficient token use, and Cognition’s Scott Wu, who cautions that ranking engineers by token spend misses true productivity. The debate reflects a broader industry conversation about how much to invest in AI inference as models become more powerful for development, research, and business operations.