Y Combinator CEO Garry Tan has revealed that he spends between $100,000 and $200,000 annually on artificial intelligence tokens, highlighting the costs associated with using AI agents for professional and personal tasks.
Speaking on an upcoming episode of Walter Isaacson's podcast, The Builders, Tan discussed how AI agents have become part of his daily workflow, supporting activities ranging from software development and research to personal information management.
Tan said his annual expenditure on AI tokens could reach $200,000, reflecting his extensive use of agent-based systems across multiple applications.
AI tokens are units of text and other information processed by AI models. Developers and users accessing these models through paid services are typically charged according to the volume of tokens processed, with costs varying by model and service.
Tan uses AI agents for coding projects, research and other professional activities. He has also made a collection of 23 Claude Code skills, known as gstack, publicly available on GitHub to demonstrate how AI agents can support different workflows.
According to Tan, his AI systems have access to information from his emails, Dropbox files and Apple Notes, including personal records accumulated over several years.
He described personal AI agents as systems capable of understanding a user's context and assisting with everyday decisions, rather than responding only to individual prompts.
Tan has also adopted OpenClaw, an AI agent platform, and believes that users should approach agent-based systems differently from conventional software applications.
His comments extended to Y Combinator's internal adoption of AI.
The startup accelerator has developed an internal AI agent called Stray Light, which Tan said contains organisational context covering founders backed by Y Combinator, applications and previous communications.
He described the system as part of an effort to change how the organisation manages information and internal operations.
The use of AI agents across large information repositories also raises questions about data access, permissions and the safeguards required when automated systems handle confidential material.
While Tan supports wider adoption of personal AI agents, he acknowledged that their current operating costs remain a significant limitation.
He said the amount he spends is not affordable for most users and that pricing would need to decline before similar systems could become widely accessible.
Tan also discussed the importance of allowing individuals to control and configure their own AI agents, rather than relying entirely on systems managed by large technology companies.
His remarks come as AI developers continue to expand agent-based capabilities that can perform multi-step tasks, retrieve information and interact with software applications with varying degrees of autonomy.
However, the cost of sustained model usage remains an important consideration for businesses and individuals evaluating these systems.
Tan's reported expenditure reflects his personal usage and should not be interpreted as a typical annual cost for AI agent adoption.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.