When should AI be allowed to approve a pull request?
Quentin Rousseau · 17 November 2026
Read the originalRootly's chief technology officer found that 90% of recent pull requests were approved without a human comment, and proposes letting an AI reviewer approve only low-risk changes.
Quentin Rousseau is chief technology officer of Rootly, which makes incident management software. As AI agents came to write most of Rootly's code, he analysed 3,661 pull requests merged between August and November 2026 to see what human approval still provided. He then proposed rules under which Rootly's AI reviewer could approve some changes without a person.
- Monthly merged pull requests in Rootly's main repository rose from 356 in October 2025 to 2,053 in October 2026, with only two more engineers approving them.
- Of the 3,661 pull requests, 90.5% were approved without a human review comment, and 43.3% were approved on an earlier commit than the one merged.
- 69% of risk labels were set under the author's account, mostly by the coding agent, and the AI reviewer rated risk higher in about 22% of paired cases.
- The proposal allows AI approval only for low-risk changes with the top score on the latest commit, never for migrations, sign-in, billing or deployment settings.
An engineering team can audit its own approval record before deciding where an AI reviewer may approve; this is an inference and is not the author's claim. Rootly has not yet switched the policy on, so its effect is untested.
Make every new commit cancel earlier approvals, so each merged pull request carries an approval of its final version.
Keep changes to migrations, sign-in, permissions, billing and deployment settings under human review whatever score an AI reviewer gives them.
Derived by Working Surface from the article; more in the Playbook. Source line: The historical analysis gives me reasons to test this rule, not enough evidence to turn it on.
19 November 2026: as AI agents write most of a team's code, the human check moves from reading each pull request to owning the rules on which changes a person reads.