I don't think so.

First of all, let's put ethics aside, that's a whole different underlying can of worms.

Now the models themselves. AIs have gotten very good at producing things we already do a lot, so "write me a bubble sort in language x". AIs are ok in medium-sized projects: e.g. "a webserver, with authentication and these sides with different UI elements etc.". But AIs are pretty bad at more complex, advanced or niche projects, e.g. "this software needs to create a VPN mesh with these other hosts, implemented control messages and dashboard, ...". Because this is not standard practice, the AI will struggle hard.

Not to go into detail on this but inherently AIs are trained in a statistical way that always introduces such asymmetric knowledge deficits.

On top of that AIs are extremely bottlenecked by lack of coherence. There's basically a window of coherence within a given conversation or project but that window always closes at some point. And the result is the AI writing mismatched code, unstructured patchwork projects and hallucinating code snippets or files. So you will get pretty far with your projects at some point but AI will always break down when you're far enough into the project. And that's when you have to have someone on board who genuinely knows the codebase.

What I'm saying is, unless we get some more fundamental changes, I have a hard time believing we get either actual AGI (and not something imitating AGI) or AI that is good enough to genuinely replace broad swaths of human jobs in any sector.