Our PRs were going through AI code review like everyone else's. Every PR came back with a long list of comments: naming, style, a missing null check, "consider extracting this function". We read all of it, fixed what made sense, and merged.
Then some of those reviewed, approved, "looks good" changes still broke in production.
The review wasn't wrong. It just answered a different question. It told us whether the code looked right. It never told us whether this change, in this part of the codebase, would hold up once it shipped.
I asked an AI assistant to review a repo before shipping it.
It came back with a clean report, a confident tone, and a long list of passes. Then I checked one of the lines it cited.
The line said something else. The mistake itself didn't bother me that much. What bothered me was that there was no shape in the output for "I don't know."
Everything was either a problem or a pass. And the passes were doing a lot of unearned work.
AI now writes 30-60% of code at modern engineering teams, but the same fragile files keep causing the same incidents, and nobody's tracking it. Tomosu scores every PR against your repo's own incident history (not generic lint rules) so you catch risk before it ships. Free self-hosted edition connects to your real GitHub repos in under 10 minutes, no credit card, no sales call. Built by a former Oracle/Hortonworks support engineering lead who got tired of watching incidents recur.
