Let agents review code so people only check risky changes
The Claude Code team says merged changes more than doubled from June to September while reverts fell over 40%, after moving most code reading to Claude reviewers.
Done with Claude Code
What they did
Anthropic's Claude Code team rebuilt its code review around agents. Claude runs tests and builds, then a fresh agent reviews the change before anyone opens it. On each change, several agents hunt for bugs and others vote on each finding. A separate security pass runs, and a classifier sets how many people must approve.
How it went
The team reports merged changes more than doubled from June to September, while reverts fell by more than 40 percent. Most changes now merge with at most one human involved.
Worth knowing
They stress giving Claude a way to check its own work, like tests or a screenshot. Repeated reviewer complaints are fixed at the source in project instruction files.
Try it yourself with Claude Code
Before I open a pull request for [repo], run the tests, build and linter and show me the output, not a summary. Then review the full change as a fresh reviewer, using [CLAUDE.md or review guide]. List each problem with a file and line as evidence, ranked by severity, starting with a count. Say plainly if there is nothing to fix. Suggest fixes, but wait for my approval before changing code or pushing anything.
Swap the highlighted [parts] for your own details.
Source: Hacker News · Addy Osmani, Morgan Westlee Lunt, Qing Wang · Oct 15, 2026
Written with AI assistance from the original post and checked against our editorial guidelines. Results are as reported by the source, not independently verified.
