Grok Bot team shares how it shipped 11,000 PRs without quality dropping

Great talk from @poteto on how the Grok Bot team shipped 11,000 PRs last month without the quality dropping:
• Reliable agents come from the codebase and checks around them, not from better prompts. Most effort goes to clean code and architecture, then lint/types/tests/verification, then a little prose in AGENTS.md.
• Every time you correct an agent, turn the correction into a deterministic check instead of another rule in markdown.
• Agents copy whatever's already in your codebase, so tech debt and workarounds spread fast. Keep it clean.
• Ban patterns that need nuance to use well. They banned useEffect in product code and about 99% of code comments, since agents use comments to justify shortcuts.
• Code review turns into reviewing proof: videos, screenshots, diagrams, and formal checks.
• Agents make thorough verification affordable. Most of her tokens now go to it: runtime checks, TLA+ for user flows, Lean for algorithms, property-based tests for UI.
• Fix the inner loop before you build a "software factory." Automating sloppy output just gets you more of it, faster.
• Their results: about one stable release a day, 227 contributors in October, a ~0.5% revert rate, and an app that got faster.