All three initials of "[A]rtificial [G]eneral [I]ntelligence" have multiple meanings.
For the sake of employment, consider what happens if we can show that some model is equal-or-better than an IQ 85 human on all tasks for which you can actually get paid, while on hardware such that it is at least real-time and costs no more than whatever minimum wage is where you are. At this point, 16% of the population are never, ever, worth paying for.
I think current models and cheapest hardware aren't quite that combination of generality and cost — but the best models can do it, and hardware of fixed performance is still getting cheaper.
1) From what I've heard*, IQ 85 is about the lowest you can be and still function as a plumber. My statement stands about everyone less clever than the worst — not average, worst — plumber.
2) for all values of X in the last two years, every time someone suggests "robots/AI are nowhere near doing X", if X is coherent**, either an AI can already do X, or at most ends up doing it within six months.
* Which is really vague due to (1) how mediocre IQ tests are in practice and (2) the internet being full of oft-repeated nonsense since day zero
** The two things people end up correctly saying AI can't do, are (1) "solve the halting problem", and (2) reinvent literally everything starting from first principles.
Capital doesn't pay labor unless it has to.
Average wages = f(labor productivity, demand for labor, labor supply)
The AGI future is that demand for labor crashes.