Eighteen months ago, I wouldn’t have guessed we’d be removing Peer Review on pull requests.
Over the last several months I’ve watched this team spend increasing effort trying to get pull requests reviewed. Not because people were avoiding reviews. Quite the opposite. They reminded each other in Slack, randomly assigned reviewers to beat the bystander effect, coordinated schedules, built skills and nudged each other throughout the day. Eventually I realized what this optimization of the review process was telling me.
I run Developer Experience at CloudZero. My job is finding friction in how engineers work, and the friction here is hard to ignore.
AI-assisted development has dramatically increased how quickly this team can produce high-quality changes. This team spent the last 6 months rebuilding the web platform. They invested heavily in the architecture, component model, testing, and the conventions that keep the codebase consistent. The review process hasn’t changed to match.
When I look at their reviews now, I see something different than I did a year ago. Most look like rubber stamps. Very few substantive comments. Its not a quality problem. AI review catches a lot before another engineer opens the PR, the automated tests do their job and the engineers understand the system deeply because they built it.
If another human is almost never changing the outcome, why are we making every pull request wait for them?
So we’re running an experiment. For code owners working in repositories they know deeply, we’re removing the requirement for a second human approval.
The engineer making the change is still responsible for it. The automated tests still run. AI review still runs. Security checks still run. We’re testing whether another required human approval is still buying enough value to justify the time it costs.
We’ll measure it the way we measure everything: pull request cycle time, throughput, and quality. If quality drops, we’ll know. If it doesn’t, we’ve removed friction without increasing risk.
I don’t think every team should do this. But every team should ask whether their processes are still earning their keep, especially as the tools around them change. Most engineering organizations have more processes than they realize. I’m probably no different. But when something changes as fast as AI-assisted development has, it seems worth checking.