TL;DR
Y Combinator CEO Garry Tan said on Walter Isaacson’s podcast that he spends about $100,000 to $200,000 a year on tokens for AI agents. He described using them for coding, research, personal tasks and work at YC, while warning that current costs put this level of use out of reach for many people.
Y Combinator CEO Garry Tan said he spends about $100,000 to $200,000 a year on AI tokens, describing an intensive use of agents for coding, research and personal tasks. Tan gave the estimate in an episode of Walter Isaacson’s podcast, The Builders, scheduled to publish Tuesday, according to TechCrunch’s November 2 report.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Tan said he uses AI agents across a wide range of work projects and coding tasks. He also used an agent to research Isaacson’s podcast and prepare for questions he might face. The quoted annual spending is Tan’s estimate; the report does not provide billing records, a breakdown by service or a precise accounting period.
Tan described giving an agent access to a large amount of personal material. “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 told Isaacson. The account illustrates how his use goes beyond isolated prompts, while also raising questions about the amount and sensitivity of information being made available to an AI system.
At Y Combinator, Tan said an internal agent called Stray Light has a personality and a “company brain.” He said it has access to context on founders the accelerator has backed, people who have emailed it and applications. Tan also shares gstack, a set of 23 Claude Code skills, on GitHub, which he described as an effort to help others learn what agents can do.
The Cost of Always-On AI Agents
Tan’s estimate puts a striking price tag on an intensive form of AI use that proponents say can handle work across different tasks and draw on a user’s accumulated context. It also shows why his experience may not represent what most people can afford: Tan acknowledged that spending this much is a “crazy privilege” and said costs need to come down for personal agents to become mainstream.
His description of agents accessing email, notes and company records also points to a trade-off beyond price. Such systems may be more useful when they can draw on extensive information, but that can involve sensitive personal or business data. The report does not detail the safeguards, access controls or data-handling arrangements for Tan’s personal setup or YC’s Stray Light system.
Tan argued that agents should not be treated exactly like traditional software, saying excessive concern about repeatability and safety can limit their useful qualities. That is his view, not evidence that safety concerns are misplaced. His comments capture a live tension in agent adoption: users may value flexible, autonomous assistance while organizations still need to understand reliability, privacy and oversight.
Products Worth Considering
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
LG UltraGear 34G600A-B 34-Inch Curved WQHD Gaming Monitor, 160Hz
Logitech MX Mechanical Wireless Illuminated Keyboard, Clicky Switches
As an affiliate, we earn on qualifying purchases.
From Personal Tools to YC Operations
Tan has previously joked about experiencing “cyber psychosis” and said he sometimes barely sleeps because he is excited to work with AI agents, according to TechCrunch. In the podcast interview, he gave a concrete measure of that enthusiasm by estimating his annual token spending. The report does not establish whether the figure includes every AI service he uses or only particular tools.
Tan also said he was an early OpenClaw adopter and is now a power user. His examples range from familiar uses—research and coding help—to systems with broad access to personal information and an internal YC agent. He has published gstack, his set of Claude Code skills, as a resource for others interested in agent workflows.
Tan said he would prefer a future in which people control and program their own personal AI, rather than relying entirely on free agents controlled by major technology companies. He compared that concern with the influence a social-media company can exert through control of its algorithm. This is his view of how personal AI might develop; the interview does not identify a specific product or timetable for such a system.
“I spend about maybe $100,000 to $200,000 a year on tokens.”
— Garry Tan, speaking to Walter Isaacson on The Builders
What Tan’s Estimate Does Not Show
The reported figure is an approximate personal estimate. The available account does not say which models or providers account for the spending, how the total is calculated, whether it fluctuates from year to year, or whether it includes costs beyond tokens. It is also unclear how much of the work done by Tan’s agents is autonomous and how often he reviews or corrects their output.
The report gives no independent measure of time saved, financial return or accuracy. It also does not describe how Tan’s personal data is stored or protected, or what permissions and safeguards apply to YC’s Stray Light agent. The podcast episode was scheduled to publish Tuesday, and further details may emerge when it is available.
Podcast Release and Cost Questions
The Builders episode with Tan was scheduled to publish Tuesday, following the November 2 report. The interview may provide more detail on how he calculates his token spending and how he uses agents across personal and YC work.
For now, Tan’s comments leave the broader questions open: whether AI-agent costs will fall enough for ordinary users, how organizations will manage access to sensitive information, and what standards they will use to check agent output. No specific price changes, deployment milestones or safety policies were announced in the report.
Key Questions
How much does Garry Tan say he spends on AI tokens?
Tan estimated that he spends about $100,000 to $200,000 a year. The figure was reported from his interview with Walter Isaacson and was not accompanied by an itemized bill.
What does Tan use AI agents for?
He described using agents for coding, work projects, research and personal tasks. He said an agent helped him prepare for the podcast and has access to personal material including email, Dropbox and Apple Notes.
What is Stray Light?
Tan said Stray Light is an internal Y Combinator AI agent with a personality and a company knowledge base. He said it has context about backed founders, people who have emailed YC and applications.
Did Tan say AI agents are affordable for most people?
No. Tan said his spending is a “crazy privilege” and that AI agents are too expensive right now. He argued that costs need to fall for personal agents to become mainstream.
Is Tan’s $100,000 to $200,000 figure independently verified?
The report presents it as Tan’s approximate estimate. It does not include billing documents, a provider-by-provider breakdown or an independent verification.
Source: techcrunch.com
