← sources

How do you stop reviewers from rubber stamping AI written pull requests

source: reddit (reddit:AI_Agents)published 2026-11-02 07:18ingested 2026-11-03 02:37

retained text

My team is about 8 devs and most of our code is written with claude code or cursor now. Reviews technically still happen, but I'm pretty sure a lot of approvals are people scrolling to the bottom and clicking approve because the PR "looks like it was written carefully".

We have coderabbit on every PR and it does catch the surface stuff, missing null checks, an unused import, that kind of thing. But last month 3 bugs reached production anyway, and none of them were the kind a reviewer bot would flag:

1. A retry wrapper that retried a payment call that wasn't idempotent  
2. A migration that was correct but locked a big table for 4 minutes  
3. A feature flag check that got inverted in a refactor, so the new flow went to everyone

Two humans approved each of those. When I asked, both said the code looked fine, which it did.

So I'm curious what's actually working for other teams:

\- Do you make the author explain the change in their own words before review?  
\- Smaller PR limits for agent code?  
\- Assigning one person as the "real" reviewer instead of two half reviewers?  
\- Something else?

War stories welcome, especially from teams where most code is AI written now

document fetched 2026-11-03 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xlqpsq/how_do_you_stop_reviewers_from_rubber_stamping_ai