A new paper on arXiv studies something that sounds small and is not: how AI coding agents write their pull request descriptions. The researchers compared five agents on the structure of their descriptions and then on what reviewers did with them: how much reviewers engaged, how fast they responded, the tone, and whether the change was merged.
The finding is that the agents have distinct description styles, and the styles are associated with differences in reviewer engagement, response time, and merge outcomes.
I think any engineering manager recognizes this, because it is true of people too. A pull request with a clear description of what changed and why gets a better, faster review than a wall of diff with a one-line title. The description is the reviewer's starting point, and its quality decides where the reviewer's attention goes.
My view is that this is a standards question, not a tool question. I think an agent should be held to the same one, and configured to produce it, because reviewer time is the scarce resource. If an agent cannot explain its change in the team's format, that is a reason to look harder at the change.
Does your team hold an agent's pull request to the same description standard as a person's?
Photo source: https://photos.robertstowe.com/wieliczka-salt-mine