2 min read
Add as a preferred source on Google

Y Combinator CEO Garry Tan Says AI Tokens Cost Up to $200,000 Annually

Tan said he uses agents for coding, research and personal tasks, and believes their costs must fall for wider adoption.

Hands pause beside an open laptop and a smartphone / TokenPost.ai
Hands pause beside an open laptop and a smartphone / TokenPost.ai

Y Combinator CEO Garry Tan said his AI-agent use costs roughly $100,000 to $200,000 a year in token fees, a personal expense he sees as a limit on wider adoption.

Tan said he turns to agents for coding, research and work projects, along with tasks from his personal life. An agent helped him research a podcast conversation with Walter Isaacson and prepare for questions that might come up. He described this kind of support as extending to routine activities, with the system able to suggest next steps.

Tan said an agent had read his email, files in Dropbox and material in Apple Notes. Among those notes were therapy records dating back two decades. He described agents as personal helpers that retain context and can offer suggestions, with some ability to act without waiting for specific instructions.

Tan has published 23 Claude Code skills under the name gstack on GitHub, an effort he said is meant to show others what agents can do.

Y Combinator is also using an internal agent called Stray Light, Tan said, as the startup accelerator changes how it operates. He described the system as having its own personality and a company knowledge base containing context on founders backed by Y Combinator, people who have emailed the organization and its applications.

Tan argued that developers should not judge agents by traditional software standards alone. He said concerns about repeatability and safety can lead users to give up some of the systems’ human-like capabilities. He described agents as tools that can take initiative as part of a person’s work.

Tan said personal agents may become common once costs drop. He described his own yearly bill as a privilege beyond what many people can spend and said the technology is too expensive now. In a separate discussion of Meta’s role in social media, Tan criticized the influence of platform-controlled algorithms and described a future personal AI that users could program and control.

The podcast conversation is scheduled for release Tuesday, Nov. 3.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…