I think AI will fall squarely into the category of the rest of the post-PC era "tech advancement" in terms of the Productivity Paradox. Outside of very niche areas (like software engineering, making software that likely wouldn't actually improve productivity in a measurable (as in, shows up in productivity statistics) way any way), AI is likely to be more of the same. All the areas where there was real benefit to digitalization and all the low hanging fruit was already finished by the mainframe and minicomputer eras.