AI code review: Are more comments actually

we’ve been thinking about something that seems easy to overlook with AI-assisted code review.

Imagine a PR gets 24 AI-generated comments.

That sounds thorough—but how many of those comments would you actually want a developer to spend time on?

There’s a big difference between:

• “This could cause a production bug.”
• “This might create a security issue.”
• “This code could be more readable.”
• “You could rename this variable.”
• “Consider refactoring this small section.”

All of these may technically be valid observations, but they don't have the same value.

For developers, the real challenge isn't just finding issues anymore.

It's prioritizing them.

Too many low-value comments can create review fatigue, hide important findings, and slow down the person who actually needs to make the decision.

I'm curious how other developers handle this:

When using AI for code review, what do you prefer?

1. AI should report everything and let developers filter it.
2. AI should only report high-confidence, high-impact issues.
3. AI should report everything but rank comments by severity.
4. Something else?

We've been exploring this problem at Codzee and believe that AI code review should optimize for signal, not comment count.

Would love to hear how this works in your team and what you've found useful—or annoying—about AI code review.
www.codzee.io