Grok Bot team shipped 11,000 PRs in a month: reliable agents come from checks, not prompts
sophiamyang · x · 2026-11-12
poteto shared how the Grok Bot team shipped 11,000 PRs last month without quality slipping:
- Reliable agents come from the codebase and checks around them, not better prompts. Priority: clean code and architecture, then lint/types/tests/verification, then a little prose in AGENTS.md.
- Turn every agent correction into a deterministic check instead of another markdown rule.
- Agents copy whatever's in the codebase, so tech debt and workarounds spread fast—keep it clean.
- Ban patterns needing nuance: they banned useEffect in product code and 99% of code comments, since agents use comments to justify shortcuts.
- Code review becomes reviewing proof: videos, screenshots, diagrams, formal checks. Agents make thorough verification affordable; most of her tokens now go to it.
More from coding & agent
- Emulating Humanlike Mouse Moves and Key Timing to Make Agents Undetectable — Darpinian · 2026-11-12
- Stanford NLP course's best homework: build your own agent harness — stanfordnlp · 2026-11-12
- Anthropic details agent memory: dreaming organizes Claude's memory, wiki+search hits 61% — RLanceMartin · 2026-11-12
- Talk preview: scaling 1,000 Fable 5.2 agents to formalize Fermat's Last Theorem — RLanceMartin · 2026-11-12
- Why Real Consumer PCs Are the Only Unblockable Way to Run Computer Use Agents — Darpinian · 2026-11-12
- Opinion: AI infra is becoming a coding runtime as intelligence units outgrow tokens — prateekj · 2026-11-12