We kept shipping bugs that passed PR + CI — so we built something to stop it
For a while, we thought our process was solid:
- PR review ✅
- CI passing ✅
- tests green ✅
And still… things were breaking in production.
Not obvious bugs.
More like:
- edge cases in async flows
- assumptions that didn’t hold under real load
- logic that “looked correct” but wasn’t safe
The frustrating part:
nothing technically failed during review or CI.
So everything felt safe — until it wasn’t.
After seeing this pattern multiple times, we realized:
CI checks correctness.
Reviews catch obvious issues.
But there’s a gap where “risky but valid-looking” code slips through.
That’s what led us to build MergAI — something that sits right before merge and tries to catch these “looks fine but risky” changes.
Still early, but already seeing interesting patterns in how PRs behave.
Curious how others deal with this:
Have you ever shipped something that passed review + CI but still broke production?
What caused it?
If you want to take a look, it’s here:
Replies