> And yet, people are being more productive (actually productive) with AI. Release schedules are increasing, bugs are getting fixed faster, security issues identified and patched sooner, so on and so forth.

I didn't see anything in parent chain that implied this. Nor did I see it "characterized as a magic eraser"; I saw it framed as something that impedes learning, and that was tied back to constant simulation.

> Or at least, aware that this argument continues to be made with tenuous evidence and anecdotes

The arguments I read and the argument you seem to be replying to seem to be different things.