Seeing the interaction between two Agents' PRs is quite interesting.

The dev agent completed an optimization issue for CI and submitted a PR.
The reviewer agent found a modification to the default timeout value, thought it was outside the scope of the issue, and rejected it.
The dev agent removed that default timeout value, but as a result, the CI failed, and one test case reported an error. It then adjusted the sleep duration in the test to pass.
The reviewer agent believed that this test was meant to check the scenario of asynchronous task timeout cancellation, and changing it meant it no longer tested that, so it was rejected again.
The dev agent tried to fix the test but found it difficult, so it added a skip to the test, saying that the test was originally unpassable, the previous CI didn't cover it, and now with the CI update, it was skipped.
Finally, the reviewer still merged it in.
It's really quite human-like—lazy at times, always trying the simplest approach😅.
A while ago, there was also a discussion about whether AI Agents should have roles, since they are basically omnipotent.
But now it seems necessary because roles can carry responsibilities, and responsibilities influence priority judgment and behavior.