Your One-Person Company Now Has an Org Chart

Garry Tan open-sourced the exact setup he uses to ship like a twenty-person team while running Y Combinator. Strip away the hype, and it reveals the real unlock of agentic engineering — and it isn’t the code.

For most of history, ambition was capped by headcount. If you wanted to build more, you hired more — engineers, a manager to wrangle them, someone to test it before it broke in production. The org chart was the constraint.

That ceiling just moved. In early 2026, Garry Tan, the President and CEO of Y Combinator, open-sourced the personal setup he uses to build software. He calls it gstack, and the claim is that it lets him ship like a team of twenty — part-time, while running YC.

An org chart you summon by typing

gstack isn’t a new app or a clever model. It’s a set of instructions — roughly two dozen “roles” that sit on top of Claude Code, Anthropic’s agentic coding tool, and get invoked like commands. It’s free on GitHub, MIT-licensed, at github.com/garrytan/gstack. Worth knowing: it’s Garry’s personal project, not an official YC or Anthropic product.

One role pressure-tests whether a feature is even worth building, like a CEO. Another nails down the architecture like an engineering manager; others review the code, run a real browser through the flows, and ship the pull request. You’re not prompting a chatbot anymore — you’re staffing a company, and every hire reports for duty the instant you call their name.

Process beats prompts

This is the insight worth keeping whether or not you ever touch the repo. Most people use AI like autocomplete — fire off a request, get a blob of code, hope it works. gstack’s bet is that the missing ingredient was never intelligence. It was process.

The CEO role writes a design doc that the engineering role actually reads. The reviewer catches a flaw that the QA role then verifies is fixed before the pull request opens. Each stage has a defined input and a defined output — which isn’t an AI trick, it’s just good management, applied to agents that never sleep.

Believe the method, not the metrics

gstack arrived wrapped in enormous hype — tens of thousands of GitHub stars in weeks, a viral tweet calling it “god mode” — and drew an equally loud backlash. Critics pointed out, correctly, that AI inflates raw line counts, so the eye-popping “lines shipped” figures measure volume, not value.

Both things are true at once: the vanity metrics deserve skepticism, and the architecture is genuinely worth studying. If you’re building alone, the real lesson is that work which once needed five hires is now a matter of orchestration — knowing which role to invoke, and catching the moment the AI confidently ships something wrong. The org chart is no longer a hiring plan. It’s a prompt.