Signals · Leverage
Garry Tan spends up to $200,000 a year on AI agents, and the money buys context
Garry Tan's agent spend shows where leverage now comes from: an agent that holds your full context and answers to you, which is still too expensive for most.
by Jo·3 min read·
New here? Start with the free AI Survival Kit →
Garry Tan spends up to $200,000 a year on AI agents, and the money buys context
0:00 / 5:05
This voice is generated by AI.
Part of the guide Your employer is grading your AI use. What to do before the next review
TechCrunch reported on 2 November that Y Combinator CEO Garry Tan spends "about maybe $100[,000] to $200,000 a year on tokens". Tan gave the figure to Walter Isaacson on an episode of Isaacson's podcast The Builders, which TechCrunch said would be published on 3 November. Julie Bort's report sets out what he gets for the money. Tan uses agents for work projects and coding. He has shared gstack, his set of 23 Claude Code skills, on GitHub. His agent also helps run his life. The headline number reads like a founder's indulgence. The detail underneath it says more about where leverage comes from in 2026.
The agent holds his context
Tan said his agent has "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 describes it as "a Jiminy Cricket on your shoulder at all times" that "will actually sort of run ahead and say, 'What about this?'"
The same logic now runs Y Combinator itself. Tan said YC has an internal agent named Stray Light with "a company brain" holding "the total context of every founder we've ever backed, anyone who's ever emailed us, like all the applications."
Read those two quotes together. The tokens are the expensive part. What Tan describes as useful is an agent that has read his email, his Dropbox and 20 years of therapy notes. AI does the routine work. Tan still decides what to do with what it raises.
Tan also argued that some people worry so much about "repeatability and safety" that they give up what makes agents useful. That is an owner's call. He runs the organisation and carries the consequences. An employee who copies the approach without the same authority takes on the risk and gets none of the cover.
Who controls the agent
Tan named the constraint himself: "It's too expensive right now." He called his spend "a crazy privilege that not everyone can do." TechCrunch suggests free agents run by large platforms may not be the fix either. On Meta he said, "This is a huge company that controls the algorithm, and we all wish that they didn't at some level," and proposed instead "a personal AI that you control, that you program."
This is a power story. The agent that knows you best will shape your attention and your decisions. Either you direct it or someone else's business model does. Right now, full-context agents belong to people who can pay for them. Everyone else gets either a limited version or one built to serve a platform's goals.
Your Next Move
-
Price one task, measure the result. Pick one recurring piece of work. Run it through an agent for a month and log the hours saved and the quality of the output against what you spent. A token bill measures usage. Your review will ask about impact, and our breakdown of how employers now grade AI use and how to show impact before your next review covers what to bring.
-
Decide what your agent can read before the tool decides for you. Tan gave his agent his therapy notes. Draw your own line on paper: personal context it can hold, work data your employer's policy covers, and material that never goes in. Context is where the value comes from, so share it on purpose.
-
Study gstack before you build from scratch. Tan published his 23 Claude Code skills so others can see what agents can do. Read how he breaks his work into skills. Then write the first three skills for your own role, starting with the tasks you repeat every week.
I build systems like the one publishing this site. → Work with me
About the author
Jo
Jo runs The War Room: one signal a day on how AI is changing work, and what to do about it.
Sources
Get the Briefing
Want more intelligence like this?
The weekly briefing, free — and the AI Survival Kit with it.
More in Signals
See allSignals · Capability
Siemens CEO Roland Busch says AI could transform society in 7 to 8 years, and AI skills will decide who gains
3 min read
Signals · Power
Windows 11 is recording which AI coding agents you run, and in one US test the upload setting was on
3 min read
Signals · Labour Market
Pay moves to the bottleneck. A careers list that is really a map of who owns the work
3 min read