The idea here is that it won't stop at 25%. Even if you were to accept this premise, maybe you're just thinking about chat gpt 3.5 or 4. But it really doesn't take a lot more imagination to think about what version 7 or 30 might do.
The same goes for the image/video generation models. Smaller production studios might forgo several artist hires and just generate the stuff they need. Large ones will have an enormous pool of unemployed creatives and won't have to pay them much at all.
Auto-completing function bodies with stack overflow content is cool! I'm not trying to say this technology isn't doing anything. It's clearly doing something "cool". But that doesn't necessarily actually make anyone more productive. That seems like an extraordinary claim (at least based on my own small experience working with it), so I'd expect to see some extraordinary evidence.
I've seen new colleagues use co-pilot at work and it definitely increases the amount of stuff they can now figure out for themselves within a given time.
EDIT: actually, I overcommitted a little bit with "really good Wikipedia search". I can rely on Wikipedia search to not invent stuff from whole cloth and try to pass it off as results.