$200K in AI Tokens

Six figures a year on AI tokens sounds like a typo. For Garry Tan, it’s a lifestyle. The Y Combinator chief opened up on Walter Isaacson’s podcast The Builders, and his comments offer a rare look at what heavy agent use really looks like. At a Glance 1. One Agent, Many Jobs Tan runs agents across work projects, coding and interview prep. He even used one to research his podcast host. To spread the know-how, he published gstack, a free set of 23 Claude Code skills on GitHub, hoping others can see what agents can do. 2. Deep Access Is the Whole Point, and the Biggest Risk Tan says his agent has read his email, his Dropbox and his Apple Notes, including 20 years of therapy notes. His analogy is a Jiminy Cricket on his shoulder, one that knows him well enough to suggest ideas before he asks. That level of access makes the tool powerful, but it also creates a large pool of sensitive data in one place. Anyone copying this approach should weigh convenience against privacy. 3. Agents Aren’t Predictable Software Tan argues that many people treat agents like traditional programs and demand strict repeatability and safety. In his view, that approach strips away what’s valuable, because agents work more like capable humans than fixed tools. Early adopters of OpenClaw like him learned to loosen the leash. 4. A Company With a “Brain” Inside Y Combinator, an agent named Stray Light has its own personality and what Tan calls a company brain. It holds the full context of every founder YC has backed, every person who has emailed the firm and every application. He says it’s changing how the organization operates. 5. Price Is the Roadblock to Mainstream Adoption Tan believes personal agents will go mainstream, but only if costs fall. He said flatly that it’s too expensive right now and admitted his own spending is a privilege. He also doubts that agents controlled by big tech are the right answer. Pointing to how Meta controls its algorithm, he suggested people may eventually want a personal AI they program and own. What This Means for Everyday Users Questions Readers Are Asking How much does Garry Tan spend on AI? Roughly $100,000 to $200,000 a year on tokens. What is gstack? His open-source set of 23 Claude Code skills on GitHub. What is Stray Light? An internal Y Combinator agent with a personality and full context on founders and applicants. What does Tan say is holding agents back? Cost. What to Watch Next Keep an eye on token pricing trends, new open-source agent toolkits and how privacy concerns shape the way people use personal AI. Source: TechCrunch