How much does it cost to live with AI agents full time? Y Combinator CEO Garry Tan put a figure on it, saying the annual bill for tokens, the units AI companies charge for usage, runs to roughly $100,000 to $200,000. The remark came on Walter Isaacson’s podcast The Builders, according to TechCrunch, with additional coverage from Mezha and Under30CEO.
What was said. Tan told Isaacson that the spending is about maybe $100,000 to $200,000 a year on tokens. TechCrunch’s headline says up to $200,000, while the quote gives a range starting at about $100,000, and the figure is a rough estimate, not an audited number. The article does not say which agents make up the spend, how many tokens that covers, or how the figure was worked out. Outlets differ on whether the podcast aired on November 2 or November 3.
What the agents do. Tan said an agent has read all of the email and Dropbox files, as well as Apple Notes and therapy notes going back 20 years, according to TechCrunch. Tan has also shared 23 Claude Code skills, called gstack, on GitHub. Y Combinator runs an internal agent called Stray Light, which Tan described as having a company brain.
The cost caveat. Tan said it is too expensive right now and described the spending as a crazy privilege that not everyone can do. Other coverage notes that the CEO thinks personal agents will only go mainstream if prices fall. This ties into a wider debate about who can afford heavy AI use, as business data from Ramp shows spending is concentrated in a small group of tech companies.
Control versus Big Tech. Tan argued for a personal AI that you control and program, and said that agents run by large technology companies are something people wish did not exist at some level, according to TechCrunch. That view mirrors the push by startups for private, local and user-controlled agents.
The privacy angle. Giving an agent access to years of private notes and email is a major privacy decision. The article does not discuss how those files are stored or protected. Anyone considering a similar setup should check where data goes, what the provider keeps, and what an agent can do with the access it has.
Why it matters. The number is a useful marker of what a heavy user pays today. For most people and businesses, token costs are far smaller, and Ramp’s data cited elsewhere shows that median firms spend $12 to $15 per employee per month. The gap between the two shows how new and uneven AI agent use is.
What to watch. Look for falling prices, cheaper personal agents, and more details on Stray Light and gstack from Y Combinator.
Sources: TechCrunch, Mezha and Under30CEO. Figures are as stated by Tan on the podcast and reported by those outlets.
San Francisco skyline (representative image), Wikimedia Commons, CC BY-SA 4.0
