Key points
- He released gstack, a set of 23 Claude Code skills, on GitHub to demonstrate agent capabilities.
- Y Combinator uses an internal agent named Stray Light with access to all founder context.
This expenditure supports his heavy use of autonomous agents for both professional duties and personal life management.
What it does
Tan uses these agents to handle daily tasks and coding projects. He described the experience as having a persistent assistant that knows him intimately. The agents read his emails, Dropbox files, Apple Notes, and even twenty years of therapy notes. They proactively suggest actions and help him anticipate needs, functioning with a high degree of autonomy.
How to get it
Tan has published a collection of twenty-three Claude Code skills named gstack on GitHub. He stated this effort aims to teach others about agent capabilities. Y Combinator also deploys an internal agent called Stray Light, which possesses a distinct personality and access to the complete context of every founder backed by the accelerator.
Limits
Tan noted that current costs are prohibitively high for mainstream adoption. He described his spending level as a privilege not available to most users. He also warned against treating AI agents like traditional software, noting that overemphasis on repeatability and safety can diminish the human-like benefits of these systems.
The takeaway
Tan believes personal AI agents will become mainstream only if costs decrease significantly. He views the current high expense as a barrier to widespread use.
Building on this
Developers can examine the gstack repository to understand how to structure Claude Code skills for autonomous tasks. The integration of diverse data sources like emails and notes suggests a pattern for building comprehensive context windows. Tan’s experience highlights the trade-off between safety guarantees and autonomous utility, offering a reference point for teams designing agent workflows that prioritize proactive assistance over rigid repeatability.
Why it matters
- The gstack release provides a concrete example of how to extend Claude Code for specific agent behaviors.
- Tan’s spending figures illustrate the current financial barrier for high-volume personal agent usage.
- The Stray Light example demonstrates how organizations can aggregate historical context for internal agents.
Topics: AI agents, Claude Code, Y Combinator, Garry Tan, gstack
Updated
The strongest doubt
Counterpoint from the sources
Tan warned that prioritizing repeatability and safety can reduce the human-like benefits of AI agents, suggesting a trade-off in design philosophy.
For readers
Developers
- Examine the gstack repository to learn how to build autonomous Claude Code skills.
- Consider the trade-offs between deterministic safety and proactive agent behavior in your designs.
Investors
- Note the high current costs of personal AI agent usage as a barrier to mainstream adoption.
- Observe how Y Combinator uses internal agents like Stray Light to leverage historical data.
Business
- Evaluate how agents can integrate with diverse data sources like emails and notes for context.
- Assess the cost implications of high-volume token usage for enterprise agent deployments.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
What is gstack?
Gstack is a set of twenty-three Claude Code skills that Tan shared on GitHub to demonstrate agent capabilities.
What data does Tan’s agent access?
His agent reads his emails, Dropbox files, Apple Notes, and twenty years of therapy notes.
Connected stories
Reading receipt
- He released gstack, a set of 23 Claude Code skills, on GitHub to demonstrate agent capabilities.
- Y Combinator uses an internal agent named Stray Light with access to all founder context.