How Anthropic Reviews 30,000 Claude Code Changes a Day

🔥 FREE AI community https://www.skool.com/ai-developer My AI engineer community: https://aiengineer.co 👋🏼 If you're new, I'm Owain. I've been building software for 20 years as a principal AI engineer and engineering director. I run my own AI consultancy, building custom software for business owners. AI code review is how the Claude Code team at Anthropic keeps up with AI-generated code. In four months, changes landing in Anthropic's monorepo went from about 2,500 a day to more than 30,000, and around 90% of Claude Code's own code is written by AI. In this video, I break down how the Claude Code team does code review at scale, including local review with a fresh agent, a REVIEW.md file, AI review on every pull request, security review, linting and type checks, and human approval scaled by risk. I also show /code-review and /security-review catching a SQL injection I planted, and run a ticket through my Software Factory plugin from triage to review. You don't need a complex setup or loads of skills for this. What you need is the same steps running on every change, so nothing depends on you remembering to do it. I'm comfortable auto merging low risk changes now, but database, payment and infrastructure changes still get human review, and no review process catches everything, so you still need a fast way to roll back. Chapters: 00:00 How Anthropic reviews 30,000 changes a day 00:42 How the Claude Code team does code review 01:21 The theory of constraints 02:21 The two jobs of code review 03:21 Review by risk and blast radius 04:16 Measuring your review process 04:45 The review layers Anthropic uses 06:04 Step 1: Local review with a fresh agent 07:06 The REVIEW.md file 08:53 Step 2: AI review on every pull request 09:16 Demo: /code-review and /security-review 10:40 Security review results 11:27 The factory mindset 13:25 One review command with Software Factory 14:28 Demo: running a ticket through the factory 15:41 Triage a rough task into a spec 16:26 Review against acceptance checks 17:03 Step 3: Security review 17:45 Step 4: Linting, type checking and build checks 18:24 Step 5: Human approval scaled by risk 19:25 Steps 6 and 7: Gradual release and fast rollback 20:48 Keeping humans in the loop 21:35 Demo: HTML explainers for code changes 23:07 Recap 24:33 Quality comes from iteration 🔗 Links How the Claude Code team does code review: https://claude.dev/blog/solving-code-... Drop a comment with any topics you'd like me to cover in future videos. #ClaudeCode #AICodeReview #AIEngineering #Anthropic