Much of the history of industrial automation has had a dichotomy between the benefit to the general economy (good) vs. the people working in specific industries that were automated (bad); only a subset of goods see total demand rise as a result of efficiency improvements.

AI is automating a lot of things at the same time. Not evenly, there are faster- and slower-automated things. But this does suggest that instead of e.g. society as a whole doing better from all the fabric coming out of the power looms, by a large enough margin to stomp on the weavers who lost their profession, that we may face something more like the entire initial economic displacement from the first industrial revolution (including the period where weavers were replaced), everything from farms finding they no longer needed so many hands through the luddites getting the death penalty to (roughly) the invention of Communism concurrently with the Irish Potato famine, and as part of this those now-bulging cities having to invent new sanitation solutions because their "slop" problem was more literal and gave the population cholera and typhoid.

My expectation a decade ago was that software development would be the last thing to be automated, because we'd need software developers to write AI. Today, while I don't know how far most fields are from automation*, we definitely have AI capable enough at coding to write more AI.

* a decade ago I also thought self-driving cars were basically already sorted; what a pity that the software in the Paint It Black video was not as good as depicted