AI-assisted code review: what actually works

Most teams adopt AI tooling at the autocomplete layer and stop there. That's fine, but it leaves most of the value on the table.

The review loop is where I've found the biggest leverage. Not because AI catches bugs better than humans — it doesn't, reliably — but because it catches categories of bugs fast enough that human reviewers can spend their attention on the things that actually require judgment.

Here's what's worked and what hasn't.