Very interesting (and... obvious) point from internal discussion: rate of being better at CONVINCING it produces Good Code outpaces rate of ACTUALLY improving, by AI coding tools.

Meaning it gets harder to spot actual problems faster than actual problems are reduced.