I feel like my role is rather to say "yes". Claude wants to keep all that legacy crap around and I keep saying shut up and purge.

I have found this to be true as well with modern claude, and to some extent codex. The blog summarized it perfectly — perhaps as a result of so much RL on security/adversarial review, the models currently will speculate a (valid, but unlikely/out of our design space/deep edge case) problem, and then cheerfully triple the codebase size and cut speed by 10x in order to fix it.

It feels exactly like talking to an experienced security engineer suddenly tasked with designing a large system — right now the tradeoffs that an architect makes don’t tend to be there, so quality oversight gives much better outcomes.