the separation between who writes and who reviews is the whole thing. I do same at smaller scale — one model writes code, different model reviews it. self-review misses things, same reason you don't review your own PRs