Code Review When AI Writes Everything

Building Claude Code with Boris Cherny

Topic 6 of 12

Published Mar 4, 2026 · Clip 40:28–47:18 (6:51)

Boris describes his old review method of logging issues in a spreadsheet and writing lint rules for repeated patterns, embodying his principle of automating himself away. Today Claude Code runs tests locally, launches itself in a subprocess to verify itself end-to-end, and every Anthropic PR gets an automated Claude code review via the agent SDK in CI, catching ~80% of bugs—with an engineer still doing the second pass and approving. He discusses deterministic safeguards (linters, type checkers, builds), best-of-N parallel dedup agents to reduce false positives, and the open-source code review skill.