AI in 2030: Extrapolating current trends [pdf]
epoch.ai
epoch.ai
The first is the difference between capability and capacity - the ability to do someinthing, versus how much of it you can do. The transition from one to the other always seems blurred in reports of this sort. That transition is beset with all sorts of phenomena that are subject to Amdahl's Law (generalised to social processes). This always makes timeframes for general availability look implausible to me.
There is a large and venerable literature in economics on the diffusion of innovations; it appears that it is being totally ignored and that the report writers are failing to distinguish between invention, innovation, diffusion, and mass deployment. (It's reasonable to ignore maturation and decay/replacement, in this case. Probably.) Diffusion and mass deployment are probably the most difficult phases (because they are intrinsically social and political), and the least discussed.
The other thing is the microeconomic focus on productivity. If the claim is made that the whole economy will be affected, then the whole economy is the unit of analysis, and the tools of macroeconomics, not those of microeconomics, are appropriate for analysis. Short run, medium run and long run phenomena are salient as well, but the reports I have seen talk as though we are instantly in the long run. I'm in my 60s and I'm still waiting to live in the long run.