n2q’s Posts
Log in
EZPost LogoPowered by EZPost© 2026 n2q
gstack: Garry Tan ships code 810x faster with 23 AI agents
n2q’s PostsAI Agents & Coding Tools
AI Agents & Coding Tools

gstack: Garry Tan ships code 810x faster with 23 AI agents

The CEO of Y Combinator claims he ships code 810 times faster than he did in 2013 — not by typing faster, but by running 23 AI agents in parallel. The tool he uses is open source.

N
Written byn2q
02 Aug 20260 min read3 views

Table of Contents

  • What it is
  • Why it matters
  • How it works
  • Caveats
  • Who it's for

#gstack: Garry Tan ships code 810x faster with 23 AI agents

The CEO of Y Combinator claims he ships code 810 times faster than he did in 2013 — not by typing faster, but by running 23 AI agents in parallel. The tool he uses is open source.

#What it is

gstack is an open source software factory for Claude Code. It packages 23 specialist skills and 8 power tools as slash commands, turning a single coding agent into a structured team: a CEO that reviews product, an engineering manager that locks architecture, a designer that catches AI slop, a reviewer that finds bugs, a QA lead that opens a real browser, and a security role that runs OWASP scans.

The framing is deliberate. gstack is not a magic button and it does not replace understanding code. It is a workflow enforcement layer — a way to make AI coding follow a disciplined sprint instead of generating undifferentiated slop. Garry Tan reportedly shipped 3 production services and over 40 features in 60 days, part-time, using this approach.

#Why it matters

  • One person can operate like a team of 20, provided they know the tools. The sprint structure — think, plan, build, review, test, ship, reflect — means nothing falls through the cracks.
  • QA is the biggest unlock. The agent opens a real Chromium browser, performs real clicks, takes real screenshots, finds a bug, fixes it, and generates a regression test. That is what reportedly let Tan scale from 6 to 12 parallel workers.
  • Security runs OWASP scans, so the output is closer to production-ready than raw agent output.
  • It is MIT licensed, free, and open source. No vendor lock-in, no per-seat pricing.

#How it works

The unit of work is a sprint. It starts with /office-hours to write a design doc, moves through /plan-ceo for a CEO-level review, then /build implements, /review finds bugs, and /ship sends the PR. Each slash command invokes a specialist role with its own prompt and acceptance criteria.

The parallelism comes from a separate Conductor tool that lets you run 10 to 15 sprints at once — brainstorming an idea, reviewing a PR, QA-ing staging, implementing a feature — managed the way a CEO manages a team. The 810x headline comes from comparing 11,417 logical lines of change per day in 2026 against 14 lines per day in 2013.

#Caveats

The 810x figure measures logical code change, not raw lines of code, and it reflects Garry Tan's own methodology — not a benchmark you will reproduce by installing the tool. gstack requires a Claude Code subscription, which runs $100 or more per month, plus Bun and Git. There is a learning curve: 23 skills take time to internalize, and parallel sprints need the separate Conductor tool. It is a workflow system, not an autopilot.

#Who it's for

gstack is for developers already using Claude Code who want to impose structure on AI-assisted work. If you ship frequently, care about review and QA discipline, and are comfortable orchestrating multiple agents, the sprint model will feel familiar. If you are looking for a one-click code generator, this is not it.

gstack is best understood as discipline for AI coding — the productivity gains come from the workflow, not from magic.

Source: https://github.com/garrytan/gstack

Filed under
AI Agents & Coding Tools
Share this post
N
About the author
n2q
Sharing ideas and building in public.
View all posts
Loading comments...

Table of Contents

  • What it is
  • Why it matters
  • How it works
  • Caveats
  • Who it's for
Keep reading

More from n2q

See all
opencodex: Use Claude, Gemini, and Grok Inside Codex CLIAI Agents & Coding Tools

opencodex: Use Claude, Gemini, and Grok Inside Codex CLI

Codex CLI was built to run OpenAI models. opencodex is a lightweight local proxy that lets you route it to Claude, Gemini, Grok, DeepSeek, Ollama, or any of 40-plus providers without waiting for upstream support.

Nn2q0 min
Agent Reach: Giving AI Agents Eyes to See the Entire InternetAI Agents & Coding Tools

Agent Reach: Giving AI Agents Eyes to See the Entire Internet

An AI agent can write code, manage projects, and summarize documents, but ask it to read a Twitter thread, search Reddit, or pull a YouTube transcript, and it usually hits a wall of API requirements, access blocks, and scattered configuration. Agent Reach wants to fix exactly th…

Nn2q0 min
AI Spec Driven: Make AI Write a Spec Before CodingAI Agents & Coding Tools

AI Spec Driven: Make AI Write a Spec Before Coding

You are coding with AI agents already, but are you still throwing vague prompts at them and hoping for the best? AI Spec Driven is a working framework that separates the thinking phase from the coding phase, and it can save you from a lot of rework.

Nn2q0 min