11,000 PRs a month at 0.5% revert rate: how SpaceXSI scales AI coding agents
AI Engineer · youtube · 2026-11-14
At the AI Engineer summit, Lauren Tan (poteto) of the Grok Bot team at SpaceXSI explains how they scale AI coding agents: 11,000 PRs in one month with a revert rate around 0.5%.
Core idea: a "trust pyramid" — build software engineering foundations before the software factory:
- Enforceable architecture with Dune
- Lint rules replacing comments as the carrier of constraints
- Runtime verification and feature maps
- Formal verification with TLA+ and Lean, plus UI testing
- Spending tokens on verification rather than generation
- Cursor agents and pstack turn these foundations into reliable software factories
Framed as "All Code Is Slop (Except Mine)": in the agent era, verification — not generation — is the scarce skill.
Related event: SpaceX Engineer Shares Agent Coding Factory Methodology(2 posts)→
More from coding & agent
- AI-built anime summer festival browser game wows — code and prompts open-sourced — prasenx · 2026-11-16
- Power user runs 5,000+ agent tasks on MulticaAI over a few months — jiayuan_jy · 2026-11-16
- 18 open-source skills turn Claude into an animation studio, led by Remotion's 7.2k-star Agent Skills — blaizedsouza · 2026-11-16
- Founder tested 12 startup ideas with cold emails at once — one got 14 yeses — doooyle · 2026-11-16
- AI-driven database optimization cuts storage/IO by ~75% and speeds queries ~85% — blaizedsouza · 2026-11-16
- AI Evals Guide: Generic Metrics Fail, Top Teams Rely on Error Analysis — PawelHuryn · 2026-11-16