gstack: Y Combinator CEO Garry Tan's Open-Source Claude Code Setup That Ships Like a Team
Garry Tan runs Y Combinator full-time and still shipped 3 production services and 40+ features in 60 days, part-time, using AI agents. His setup is now open source: gstack turns Claude Code into a virtual engineering tea
Garry Tan runs Y Combinator full-time and still shipped 3 production services and 40+ features in 60 days, part-time, using AI agents. His setup is now open source: gstack turns Claude Code into a virtual engineering team, with 23 slash-command specialists acting as CEO, designer, eng manager, release manager, doc engineer and QA. All Markdown, all free, MIT licensed. 136,500 stars and counting.
The Y Combinator CEO Who Ships Like a Startup
The most interesting thing about gstack is not the code. It is the person who wrote it.
Garry Tan is the President and CEO of Y Combinator. Before that, he was one of the first engineers, PMs and designers at Palantir, co-founded Posterous (which sold to Twitter), and built Bookface, YC's internal social network. He has worked with thousands of startups, including Coinbase, Instacart and Rippling back when each was one or two people in a garage.
And in the middle of running YC full-time, he has been quietly shipping more software than at any point in his twenty-year career. gstack is his exact personal setup for doing it, the slash commands and agent skills he uses every day, open-sourced with an MIT license. His own description is blunt: "This is my open source software factory. I use it every day."
810x: The Numbers Behind the Software Factory
The README opens with a quote from Andrej Karpathy: "I don't think I've typed like a line of code probably since December, basically, which is an extremely large change." Tan heard that and wanted to find out how one person ships like a team of twenty. Part of his answer came from watching Peter Steinberger build OpenClaw, now at 247,000 stars, essentially solo with AI agents.
Then he published his own numbers. In the last 60 days before the README was written: 3 production services, 40+ shipped features, part-time, while running YC full-time. Measured on logical code change (not raw lines of code, which AI inflates), his 2026 run rate is about 810 times his 2013 pace: 11,417 versus 14 logical lines per day. Year to date, 2026 had already produced 240 times his entire 2013 output, measured across 40 public and private repos including Bookface.
He is upfront about the caveat. AI wrote most of it. His point: "The point isn't who typed it, it's what shipped." There is even a full methodology document in the repo, ON_THE_LOC_CONTROVERSY.md, with the reproduction script, for anyone who wants to argue with the measurement.
What gstack Actually Gives You
Strip away the story and gstack is this: it turns Claude Code into a virtual engineering team. A CEO who rethinks the product, an eng manager who locks the architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP and STRIDE audits, and a release engineer who ships the PR.
Concretely, that is 23 specialists plus 8 power tools, all exposed as slash commands, all Markdown, all free. The command list reads like an org chart: /office-hours, /plan-ceo-review, /plan-eng-review, /review, /qa, /ship, /cso, /autoplan, /retro, /investigate, /learn, and more.
The core loop the README recommends:
- Run
/office-hoursand describe what you are building - Run
/plan-ceo-reviewon any feature idea - Run
/reviewon any branch with changes - Run
/qaon your staging URL - Stop there. You will know if this is for you.
Installing gstack Takes About 30 Seconds
Requirements: Claude Code, Git, and Bun v1.4.2 or newer (Node.js additionally on Windows). On macOS the Aside browser is recommended for the browser-driven skills, but setup builds its own bundled browser if it is missing.
Open Claude Code and paste this. Claude does the rest:
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup
Then add a "gstack" section to your CLAUDE.md listing the available skills. There is also a team mode (./setup --team) that bootstraps a shared repo so teammates get gstack automatically, with no vendored files and no version drift.
Claude Code is the fully supported host, but gstack is not Claude-only. ./setup --host auto installs for every agent it detects, with Codex CLI, OpenCode, Cursor, Kiro and GitHub Copilot CLI at an experimental tier, and OpenClaw gets native support since it spawns Claude Code sessions.
Who It Is For
Good fit if you are a technical founder who still wants to ship, a first-time Claude Code user who wants structured roles instead of a blank prompt, or a tech lead who wants rigorous review, QA and release automation on every PR.
Poor fit if you want a fully hands-off pipeline. gstack keeps you in the loop by design: /office-hours interrogates you, /plan-ceo-review challenges your scope. If you would rather the agent just go, the structure will feel like friction.
Read the full breakdown with FAQs on DevPik: https://www.devpik.com/open-source/gstack
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.