Garry Tan open‑sources a simulated company stack
Garry Tan released 'gstack', an open‑source AI setup that simulates a virtual tech company with agents for CEO, engineering manager, QA and a conductor enforcing strategic thinking before coding shared. The project packages role‑based agent workflows as a sandbox for leadership and process experiments.
The repository garrytan/gstack on GitHub lists an initial release labeled v0.0.1 and shows roughly 2.4k stars with about 313 forks on the project page. github.com The code tree in the repo exposes six top-level SKILLs such as plan-ceo-review and plan-eng-review in the main directory, according to the repository listing. github.com Press coverage and summaries, however, enumerate eight user-facing commands — naming /plan-ceo-review, /plan-eng-review, /review, /ship, /browse, /qa, /setup-browser-cookies and /retro — highlighting a small divergence between the repo layout and external writeups. marktechpost.com gstack’s core technical design centers on a long‑lived headless Chromium daemon that preserves cookies, tabs and localStorage to reduce cold-start latency, a decision MarkTechPost documents as the project’s most important subsystem. marktechpost.com Several walkthroughs describe gstack as a local‑first Claude Code configuration that wires CLAUDE.md skill files, a Bun-based runtime and Playwright-driven browser automation into repeatable slash commands for planning, review, QA and shipping. sitepoint.com Feature-level behavior in tutorials includes a /ship command that syncs with main, runs tests and opens a PR automatically, and a /browse mode that drives the app, takes screenshots and lets the agent perform visual QA against the active branch. maven.com A paid masterclass and multiple how‑tos surfaced within 48 hours of the release, with a Maven course teaching the Brian Chesky‑style product framing baked into /plan-ceo-review and SitePoint publishing an install-and-configure walkthrough on March 13, 2026. maven.com
