Guide

Have one AI agent review another’s code

An agent that wrote the code is a poor judge of it: it tends to share its own blind spots. A second agent on a different model makes a better first reviewer. YOLO Studio builds that into the Background Runner as Agent review.

Get started

How Agent review works

Turn it on
In the Background Runner settings, set Agent review to Required. It is fixed onto each card when the card starts, so it applies to cards started afterwards.
The work finishes
When an implementation succeeds, the card stays in Doing and moves to Queued for review, then Reviewing.
A different model reviews it
The reviewer must run a different model from the one that did the work, and a different model family is preferred. If none is available, the review waits for you; a model never reviews its own work.
Approve or request changes
An approval covers exactly the commit or result it reviewed. Changes requested means nothing relies on the result until you choose Fix findings, which queues a new attempt that is reviewed again.

What the review records

The card shows the verdict, the findings, the checks the reviewer ran and their limits, the exact commit or result reviewed, and why that reviewer was chosen. Review failed means the review could not run, not that the code failed it.