Code Review

Code review is the step where, before handing off changed code to someone else, a developer takes one more look to catch mistakes or weak spots. It's similar to how an editor rereads a manuscript before publication, marking typos and logical gaps. But the editor doesn't rewrite the manuscript and send it straight to print — they only mark it up, and the author decides whether to actually print it.

The Code Review feature introduced in Claude Code has AI take on this editor's role. Once a development team turns it on, every time code changes are pushed, it automatically compares them against the full codebase and leaves comments tagged by severity on lines that could be problems. It's designed to avoid piling up duplicate remarks on similar issues, instead prioritizing them so only the few things a developer really needs to see get surfaced.

What matters is that this feature never makes the decision to approve or block a merge. Its role ends at posting the feedback — whether to actually act on that feedback and merge the code is always up to a human. In other words, a human always stands at the final gate.

Articles put it this way: "Code Review never decides to approve or block a pull request. That judgment is always left to a human." A common misconception is thinking AI review handles the merge on its own. In reality, it only leaves comments pointing out issues — acting on them and finalizing the merge is something the developer has to decide and do themselves.

If your organization uses Claude Code, try turning on the Code Review feature in the admin settings, and choose which repositories to watch and when it should run (when a pull request opens, on every push, or when requested via comment).

If it's already set up, you can leave "@claude review" in a pull request comment to check whether an automatic review gets attached.

If you want to actually apply the changes after review comments come in, try running the "/code-review" command in your terminal to review the diff again, then add the "--fix" flag to apply the fixes to your working tree.