Sounds sort of like Bracha's stance.
I think we should just focus on making the feedback better over an entire program execution. And we could make defining unit tests easy, which would provide the context needed to test smaller parts of the program.