I'm curious about integrating Jev with testing. You throw it on a loop to check the test results and have the more general AIs fix things until tests pass. I'm sure something like this exists but it now can have more determinism and cheaper.
It might be good at routing fixes to the most appropriate LLM.
I used Jev to make my AI game 6x faster and 8.7x cheaper.