Y Combinator The CEO once half-jokingly admitted that he suffered from “cyber psychosis”, and sometimes he could hardly sleep because he was too excited about working with AI agents. In a podcast on Walter Isaacson The Builders In the new episode of the program, which is scheduled to be released on Tuesday, Tan has assigned a price to their own enthusiasm.
“I spend around $100,000 to $200,000 per year on tokens,” he told Isaacson. He uses agents for countless work projects and coding tasks, and has publicly shared a set of 23 Claude Code skills on GitHub, which he named gstack; he said that this is an effort to help others understand what agents can do.
On one hand, the way he uses agent is very similar to how many people use AI: to help himself study the podcast content of Isaacson and to prepare for possible questions. On the other hand, his agent also helps him manage his life.
“It has read all my emails, all my Dropbox. It has also read all my Apple Notes, including my treatment notes from the past 20 years,” he said.
He described using AI agents as being like having help with “almost anything” in daily life, and moreover, it allows one to complete tasks more independently. “Think of it as a Jiminy Cricket always on your shoulder. It knows almost everything you want to do and will be there to help you. It might even take proactive action, saying, ‘How about this?’ Or it can help you see what’s around the corner.”
As an early adopter of OpenClaw and now a heavy user, Tan says that one thing he has learned is not to treat AI and agents as traditional software. Some people are overly concerned about “repeatability and security to the point where they actually sacrifice those wonderful aspects of collaborating with AI and agent, which can essentially be very similar to human interactions,” he said.
For example, Y Combinator is also using a agent. “We need to rebuild the way we operate in YC,” he said. “For instance, we now have an internal AI agent called Stray Light; it kind of has its own personality. It also has what’s like a ‘company brain.’ In fact, it holds all the context regarding every founder we’ve supported in the past, as well as information from anyone who has sent us emails and all application materials.”
Tan is certainly an avid tech enthusiast – after all, he is in charge of a famous startup accelerator in Silicon Valley. He believes that personal AI agents will eventually become mainstream, provided that one major issue is resolved: the cost must be reduced.
“It’s too expensive now,” he said, adding that he is aware that the money he spends on AI and agents is “some kind of crazy privilege that not everyone can afford.”
But free agent controlled by large tech companies may not necessarily be the answer. When talking about the future he hopes technology will build, and the impact of Meta as a social media company on us, he said, “This is a big company that controls algorithms, and to some extent, we all hope they wouldn’t do that.” He added, “So you can imagine a personal AI that you control yourself and program yourself.”
