The thing to keep in mind is that the march of technology has always been oriented toward reducing the number of people required to do something. Arguably the single most significant development that led to the modern 'digital age' was Guttenberg's movable type printing press, as it basically took the work of scribes painstakingly rewriting books by hand into new physical volumes and turned it into a process that could mass-produce (for the time period at least) the text of the same works in a fraction of the time with a fraction of the labor.
This had downstream effects on recordkeeping and led to the development of other tools like the typewriter, itself ultimately leading to digital records systems and the modern computing age. At each step along the way, the number of low-level clerks has reduced, and I've seen it firsthand in my own lifetime, where a department's office would have an executive secretary, an administrative secretary, and a receptionist, slowly eliminate to just a single person because so many of the clerical records duties or other secretarial duties were reduced or eliminated by technology.
Now, that said, these tasks were not typically oriented towards decision-making so much as following procedures. That's why these tasks were able to be automated or otherwise condensed or shifted to different personnel, because they were just performing rote steps that were able to be automated. AI could end up being different in that they're now trying to replace creative or conclusion-drawing tasks with it, something that may or may not work out in the end. Right now it's definitely not ready for it, and given how easy it seems to be to poison AI's fundamental dataset, I am left wondering if it ever will be, or if so many jobs will be needed to 'curate' that dataset that it draws upon that in the end it won't really save money. It'll just transfer that money to contractors instead of local staff.