Garry Tan, the chief executive of Y Combinator, has previously joked about experiencing what he calls "cyber psychosis" - a state of near-sleepless excitement driven by his enthusiasm for working with artificial intelligence agents. In a forthcoming episode of Walter Isaacson's podcast, The Builders, scheduled to go live on Tuesday, Tan moved beyond the jokes and attached actual figures to his AI habit.

"I spend about maybe $100,000 to $200,000 a year on tokens," Tan told Isaacson during the conversation. The spending reflects a remarkably deep integration of AI agents into both his professional and personal routines. He relies on agents to assist with a broad range of work projects and coding tasks, and he has publicly shared a collection of 23 Claude Code skills on GitHub under the name gstack, describing it as an educational resource to help others understand what AI agents are genuinely capable of.

An Agent That Knows Everything About Him

Tan's use of AI agents spans a wide spectrum. At one end, he uses them in ways that have become relatively familiar to many professionals - for instance, researching Isaacson's podcast ahead of the recording and anticipating the types of questions he might face. At the other end, however, his agent operates at a far more intimate level.

"It's read all of my email and all of my Dropbox. It's read, like, all of my Apple Notes, even my therapy notes going back 20 years," he said during the episode.

He frames the experience of using AI agents as something that touches nearly every aspect of daily life, describing it in vivid terms. "Almost anything that you do in your day-to-day life, and for it to do it somewhat more autonomously. Think of it as a Jiminy Cricket on your shoulder at all times. Like, almost anything you want to do, it knows you, and it's there to help. And it will actually sort of run ahead and say, 'What about this?' Or it'll help you see around corners," he explained.

Don't Treat Agents Like Conventional Software

As someone who adopted AI agent tools early and has since become a power user, Tan has formed clear views on how people tend to misuse or underutilize them. His central observation is that users should resist the temptation to treat agents the way they treat traditional software applications.

He noted that some users become so preoccupied with "repeatability and safety that they actually trade off all of the beautiful parts of working with an AI agent, which basically can be so similar to a human." In his view, this overly cautious approach strips away much of the value these tools can offer when given appropriate latitude.

Y Combinator's Internal Agent, Stray Light

Beyond his personal use, Tan revealed that Y Combinator itself has deployed an internal AI agent that is actively reshaping how the organization operates. The agent, which Tan said has been given a name and a defined personality, is called Stray Light.

"We now have an internal AI agent, for instance, named Stray Light that kind of has a personality, and then it has a company brain. Like it actually has the total context of every founder we've ever backed, anyone who's ever emailed us, like all the applications," he said. The implication is that Stray Light functions as a comprehensive institutional memory and operational assistant for one of the world's most influential startup accelerators.

The Cost Problem Standing in the Way of Mainstream Adoption

Despite his enthusiasm, Tan is candid about the single biggest barrier preventing AI agents from reaching a wider audience: the price of using them is still far too high for most people. He acknowledged this directly, noting that the amount he spends annually is "a crazy privilege that not everyone can do."

"It's too expensive right now," he said plainly. For personal AI agents to become a genuinely mainstream tool rather than a luxury available only to well-funded technologists and executives, the cost of running them needs to fall significantly. This is a point that aligns with broader industry discussions about the economics of large language model inference and the need for continued efficiency improvements.

The Case for Personally Controlled AI

Tan also used the conversation to touch on a broader philosophical question about who should ultimately control the AI agents that people come to rely on. He drew a pointed comparison to the way large social media platforms, particularly Meta, have shaped user experience through algorithmic control - often in ways that users find uncomfortable or disempowering.

"This is a huge company that controls the algorithm, and we all wish that they didn't at some level," he said. He suggested that the better path forward would involve AI that individuals own and direct themselves, rather than agents provided and governed by large technology corporations with their own commercial interests.

"So you could imagine a personal AI that you control, that you program," he added - pointing toward a future where the relationship between a person and their AI agent is defined by individual autonomy rather than platform dependency. Whether the technology and business models needed to make that vision real will materialize at accessible price points remains, for now, an open question.