Would be interesting if you honestly documented as to if/when it starts breaking old things as you add new features.

scbench.ai suggests this will happen on projects where the human is no longer in the loop, and it's a good sized project to find out.