It takes longer to write a prompt that will get me 90% of the way to what I would write myself and then rewrite and recheck everything for that last 10% than it usually takes me to just do the thing myself in almost every case.

The one thing I do find it useful for is doing executive summaries for something I've already written but even there it takes more time because I feel like the standards for polishing and polishing and polishing something (and double checking that AI hasn't dropped out something important or slipped in something untrue) takes more time and is less satisfying work than writing it myself.

I do like AI but I can't say hand on heart that it saves me any meaningful amount of time or significantly improves the quality of my work.

Where it's a nightmare is when new college grads with imposter syndrome use it for everything to sound smart and then I have to spend hours in peer review rewriting all the screeds of AI slop they produce. That significantly increases my workload because it takes twice as long and there's no learning curve when the newbie didn't write it themselves and has no intention of even trying to learn how to write it better in the future.

people who are actually experts on whatever genAI is generating, who point out that just catching and correcting its errors can eat up or exceed the time it supposedly saves, and people who aren't experts but want to pretend they are - imposter syndrome, as you said - who don't check the AI results carefully enough, and in some cases don't check at all.

There are now AI-written papers in scientific and medical journals online - papers that got past what was apparently peer review that didn't happen or was done by AI - that include what are clearly AI responses to prompts.

I posted an OP recently that referred to an article mentioning a survey of developers that found about half trusted AI enough they didn't bother to check results. It's scary to think how much risky AI code is already in use.

... as one realizes the problems they are solving have been solved thousands of times before and one, like Sisyphus, is pushing a boulder up an ever growing mountain of abstractions, only to see it roll down again before it reaches the top.

New creatures are not born and do not grow in the underworld of generative AI. It is a dead place of eternal repetition and toil.