Sometimes when I'm talking about IP law and why AI output shouldn't be copyrightable.
Sometimes when I'm talking to another human and explaining to them that yes, a real human is going to address their problem, because they're concerned about AI fucking it up.
I've just started "using" AI-generated summaries at work. Thankfully this is optional, so by "using" I really mean "mocking". They are nowhere near good enough to use without extensive revision. They get way too much wrong to be good for anything other than catching some minor points I didn't have time to note. They seem to work well for short interactions, but I have no trouble noting everything that occurred under those circumstances, so there's no value there.