This makes me think a bit about my wife - she works in a professional field, but ends up spending a bunch of her time on clerical work because it's nearly impossible to hire and keep good assistants.

When I look at the clerical work from a software engineer's point of view, I find it nuts how few guardrails there are - clerical workers would have a far easier time producing work with a low error rate if there were some guardrails put up, but that's just not an investment/culture that the field seems to be interested in.

The two related points around AI are that 1) guardrails would make the clerical work relatively easier for AI and 2) because assistants are so error-prone anyway, the bar for AI isn't all that high.

In a lot of systems the kind of rigid guardrails we have in computers are counterproductive. It descends into a epicyclical, fractal mess of special-cased exceptions, none of which faithfully models the actual system [EDIT: an example which illustrates both this problem and how following through through can--if successful!--yield interesting results is in the preface to The Structure and Interpretation of Classical Mechanics: https://groups.csail.mit.edu/mac/users/gjs/6946/sicm-html/bo...].

Some problems that seem straightforward at first blush might in fact be AGI-complete--that is they require actual judgment and reasoning to solve. I'm not making the specific claim that the clerical work you're describing is one of those, but it could take a large amount of data modeling work to determine whether it is.

This is what makes finding productive AI (EDIT: I mean ML, AI == AGI and we don't have that) applications so challenging. It's why my money is firmly not on a big AI revolution anytime soon, despite the demonstrated capability of language models.