An ambient reviewer that watches the branch and specifically targets the failure modes of AI-generated code that humans now waste hours on: diffs that touch 11 files for a 3-file change, scope creep, and tests that were 'made green enough to move on' without proving anything. It triages the PR (risky changes, fake-green tests, over-broad blast radius) so senior engineers only spend judgment on architecture, not on tracing AI output line by line. STRONGEST of the set — the bottleneck has measurably shifted from writing code to reviewing AI-written code.
Built from real signals: (4394 upvotes, 149 comments); OpenAI just made a deal with a devil and lost this customer of 2 years. The company (originally non profit) that told us; EDIT: I’ve added a small update/edit at the bottom of this post 2 EDIT: I don’t know if I’m supposed to write this here