YC shares open-source AI tools after revealing up to $200K annual spend on personal agents
YC chief details token spend on personal and corporate agents in a new podcast episode, sharing open-source tools for others.
Key points
- Garry Tan spends up to $200,000 annually on AI tokens for personal and work agents.
- YC uses an internal agent named Stray Light with full context on past founders.
- Tan shared 23 Claude Code skills as gstack on Github to help teach agent use.
The facts
His personal agent has access to his email, Dropbox, Apple Notes and therapy notes from twenty years ago. He shared his 23 Claude Code skills, called gstack, on Github.
Who it affects
This reporting concerns YC executives and the broader developer community. Tan noted that YC uses an internal agent named Stray Light that holds context on every founder backed by the accelerator and all applications received.
What is still unknown
The specific models powering these agents were not named in the report. It remains unclear how YC secures the data fed into Stray Light or how Tan manages the security of his personal notes being ingested by an AI system.
The takeaway
Tan believes agents should not be treated like traditional software. He argued that focusing too much on repeatability and safety can trade off the human-like benefits of agents. He described the experience as having a constant helper that knows the user well.
What changes for users
Users can access Tan’s gstack tools on Github. However, he emphasized that current costs are a privilege. Data science teams should note his warning that high spending is not scalable for most users yet. Costs must decrease for mainstream adoption.
Why it matters
- Tan’s admission of spending up to $200,000 annually on tokens underscores the high financial barrier currently preventing mainstream adoption of personal AI agents.
- The use of an internal agent with total context of every founder and application demonstrates how AI can fundamentally restructure organizational operations and memory.
- Tan’s warning against prioritizing repeatability and safety over human-like interaction suggests a shift in how users should approach the design and utility of AI agents.
- The integration of agents into personal life management, including access to decades of therapy notes, illustrates the depth of autonomy and contextual awareness these tools can achieve.
Topics: AI agents, Y Combinator, token costs, Claude Code, Garry Tan
Updated
Then / now / next
AI agent usage was often experimental or limited in scope.
Power users like Tan are spending heavily on comprehensive agent systems.
Watch for cost reductions that could make this level of access mainstream.
Story timeline
- Then AI agent usage was often experimental or limited in scope.
- 03 Nov 2026 Y Combinator’s Garry Tan uses AI agents so much, he spends up to $200K a year on tokens
- Now Power users like Tan are spending heavily on comprehensive agent systems.
- Next Watch for cost reductions that could make this level of access mainstream.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
What is gstack?
It is a set of 23 Claude Code skills shared by Tan on Github.
Does YC use AI agents internally?
Yes, YC uses an agent named Stray Light with context on all past founders.
Connected stories
Reading receipt
- Garry Tan spends up to $200,000 annually on AI tokens for personal and work agents.
- YC uses an internal agent named Stray Light with full context on past founders.
- Tan shared 23 Claude Code skills as gstack on Github to help teach agent use.
Still open: How does Tan secure his therapy notes when ingested by an AI agent?
Still open: What specific model providers does Tan use for his personal stack?
