AI code review: the referee has to be you

· Published · 1 min read

If you ask an AI agent to review code, there are two cases:

  • either it built the feature itself, and it will often say that, overall, everything is well done;
  • or another agent built it, and then it will systematically find something to criticise.

And you can rerun it 5, 10, 20 times if you like… something else will always come up.

In other words: the verdict depends as much on how you ask the question as on the code itself.

The referee has to be you

The solution to this thorny problem?

Be the referee yourself.

Really work on your software engineering knowledge, so you know what is truly good, bad, open to debate… and why.

Playing that referee is also what I do in my clients’ code reviews.

Happy learning!