How the Claude Code team reinvented code review with AI

Moving review from gatekeeper to guide

Addy Osmani, Morgan Westlee Lunt, and Qing Wang recently described how the Claude Code team has changed the way code review works. Instead of treating review as a final checkpoint before merging, they now treat it as a continuous process — more like spellcheck while you write than a barrier at the end of the road.

"We think 'did a human read the code?' is an imprecise proxy for 'what could go wrong with this change, and which check — automated or human — is the right one to catch it?'" the three wrote together.

That shift in thinking has reshaped their entire workflow.

Claude reviews first, humans review second

Under the new system, most code changes are reviewed and fixed by Claude at least once before any human is asked to look. The AI catches the routine mistakes — syntax errors, style issues, simple bugs — so that the team's human attention can be saved for higher-level questions and deeper scrutiny.

This approach turns out to be both faster and more effective than the old model, where every pull request sat waiting for a colleague to free up time to read it. Now the AI does the first pass automatically, and the human focus is on judgment, not correction.

Why it matters beyond Anthropic

The Claude Code team's experiment offers a practical example of how AI coding agents are changing the day-to-day life of software engineers. Rather than replacing human reviewers entirely, Claude acts as a filter — handling the volume of routine checks so that people can focus on what machines still struggle with: architecture decisions, security concerns, and whether a change actually makes sense in context.

As AI coding tools become more common, the question is no longer whether machines can read code. It is who decides what kind of review each piece of code needs, and when a human should step in.

For the Claude Code team, the answer is clear: let AI catch the small stuff first, then bring people in for the hard questions.

Key facts

• Most changes are reviewed and fixed by Claude at least once before a human looks at them
• The team treats code review as a continuous process, like spellcheck, rather than a final gate
• AI handles routine mistakes so humans can focus on higher-level questions and scrutiny
• The approach was described by Addy Osmani, Morgan Westlee Lunt, and Qing Wang

Sources

1. A quote from Addy Osmani, Morgan Westlee Lunt, and Qing Wang — simonwillison.net