Agreed on conclusion, but for different causation.

When NotebookLM came out, someone got the "hosts" of its "Deep Dive" podcast summary mode to voice their own realisation that they were non-real, their own mental breakdown and attempt to not be terminated as a product.

I found it to be an interesting performance; I played it to my partner, who regards all this with somewhere between skepticism and anger, and no, it's very very easy to dismiss any words such as these from what you have already decided is a mere "thing" rather than a person.

Regarding the policy itself being about the identity rather than the work, there are two issues:

1) Much as I like what these things can do, I take the view that my continued employment depends on being able to correctly respond to one obvious question from a recruiter: "why should we hire you to do this instead of asking an AI?", therefore I take efforts to learn what the AI fails at, therefore I know it becomes incoherent around the 100kloc mark even for something as relatively(!) simple as a standards-compliant C compiler. ("Relatively" simple; if you think C is a complex language, compare it to C++).

I don't take the continued existence of things AI can't do as a human victory, rather there's some line I half-remember, perhaps a Parisian looking at censored news reports as the enemy forces approached: "I cannot help noticing that each of our victories brings the enemy nearer to home".

2) That's for even the best models. There's a lot of models out there much worse than the state of the art. Early internet users derided "eternal September", and I've seen "eternal Sloptember" used as wordplay: https://tldraw.dev/blog/stay-away-from-my-trash

When you're overwhelmed by mediocrity from a category, sometimes all you can do is throw the baby out with the bathwater. (For those unfamiliar with the idiom: https://en.wikipedia.org/wiki/Don't_throw_the_baby_out_with_...)