How are you guys able to not review one another's PRs at all? How essential is Mendral to this process working? https://www.mendral.com/

Rust is a HUGE factor, yes. A couple other factors - we have very good test suites and a culture of running them locally - many of our PRs are for the upcoming BAML language, which isn't in prod - we often read PRs post-merge and discuss them internally Mendral isn't exactly a core factor, we only started using it a couple days ago. We've always had some PR bot or another, and occasionally they catch things. But so far they haven't provided as much value as the points above.

Answered the question I wanted to ask, always curious if AI review helps improve code

Yeah somehow I don't see this approach working well with Python codebases 😅. Rust is just amazing for this. 👌🏽

My friend is doing a CS PhD in formal methods, making proofs about graph theory with HOL4. According to him, opus-4 can write certified code much better than he can. So in the same vein as Rust complementing the coding agents, even more advanced type systems seem to be advancing fast too. The age of dependently typed coding assistants is coming, IMHO. (This is my personal opinion and does not reflect the beliefs or views of my employer BoundaryML :P)

Curious how Mendral works out. Looks like a next gen exception tracker that takes action on its own, potentially, to mitigate or fix production issues.

After just a few days, actually I'm very impressed with it. It's definitely catching some non-trivial stuff and producing way less fluff than the previous tools we've tried.