AI Explainedby Milad Fahmy

AI Explained

AI code review: a model on your pull requests

AI code review tools read your diff and comment on the changed lines. What they catch, how Copilot and Claude fit in, what Google measured, and why a person still decides.

Milad FahmyNov 11, 20262 min read

On this page
  1. How it works
  2. Copilot on GitHub
  3. What Google measured
  4. The limits
  5. What humans still review
  6. A good setup

Your next code reviewer might not be a person. It's a model, and it's probably already on your pull requests.

How it works

The model reads the diff and the code around it, then leaves comments right on the changed lines, just like a human reviewer would.

It's good at the boring catches:

  • typos
  • missed null checks
  • unused variables and inconsistent names

Copilot on GitHub

On GitHub, you assign Copilot to a pull request and it reviews it like a teammate would.

  • Many comments come with a suggested change you can apply in a couple of clicks.
  • You can turn on automatic reviews, so new pushes get checked without anyone asking.
  • You can teach it your team's rules in an instructions file: .github/copilot-instructions.md.
- Use our logger, not console.log
- Every endpoint needs a test
- Flag raw SQL

Other tools work the same way. With Claude Code's GitHub Action, you mention @claude in a pull request or issue comment to get it to look at the code.

What Google measured

Google tried a version of this at scale: a model that proposes code edits to resolve reviewers' comments.

  • Getting a change through review took authors about an hour of active work on average, between sending it for review and submitting it.
  • Authors applied 40% to 50% of the suggested edits they previewed.
  • Google expects it to save hundreds of thousands of hours a year.

The limits

AI review isn't perfect. GitHub's own docs say Copilot won't spot every problem, and sometimes it makes mistakes.

By default, Copilot's reviews don't count toward required approvals. A person still signs off.

Too many comments also turn into noise. Tune the tool to flag what matters, and leave style to your linter.

What humans still review

People review what a model can't judge:

  • the design
  • the product intent
  • whether the change should ship at all

A good setup

Step Job
Linters and tests The mechanical checks
AI pass The small stuff a person might miss
Human reviewer The call

Let AI catch the small stuff. You own the decision.

Sources

#ai code review#code review#pull requests#github copilot#developer tools

Watch