How I Review AI-Generated Code

70% of teams report worse code quality since adopting AI tools. The problem isn't AI. It's that most teams have no review process for AI-generated code.

In this video, I walk through my 4-layer approach to AI code review. Each layer catches a different category of problems so that by the time a human reviews the code, most issues are already handled.

Layer 1: Automate the obvious (hooks, linting, tests)
Layer 2: Review locally (AI reviewing AI, fresh context)
Layer 3: Automated CI checks (safety net on every PR)
Layer 4: Human review (architecture, business logic, judgement)

I demo each layer with real code and show you the exact configs I use.

All configs, hooks, and review prompts are linked below.

⏱ Timestamps
0:00 The problem with AI-generated code
0:45 Why code review matters
1:30 AI review vs human review
2:30 Layer 1: Automating the obvious
4:00 Layer 2: Reviewing locally
5:30 Layer 3: Automated platform checks
6:15 Layer 4: Human review
7:00 My honest take

🔗 Resources
Blueprint plugin (code review skill): https://github.com/owainlewis/blueprint/blob/main/skills/code-review/SKILL.md
Codex review prompt (reference): https://github.com/openai/codex/blob/main/codex-rs/core/review_prompt.md
CodeRabbit CLI: https://docs.coderabbit.ai/cli/claude-code-integration
Codex GitHub integration: https://developers.openai.com/codex/integrations/github
Anthropic security review action: https://github.com/anthropics/claude-code-security-review

📰 Newsletter: https://newsletter.owainlewis.com
💬 Community: https://skool.com/aiengineer

Need help with a project? Work with me: https://gradientwork.com.

#AIEngineering #CodeReview #ClaudeCode #SoftwareEngineering