Two repositories, the same question.

How much more of the human review loop did the pull requests carrying agent authorship take — in each of them?

Or pick two from everything already read.