some of these genuinely excite me, like the slate recognition (chore-reduction), clip search (although I'd want to see how reliable it is), and deblur (a typical PITA post fix). anything that makes masking, tracking, and level-matching easier saves hours on hours, but only if they're either reliable or easy to fix what the automation gets wrong (and it'll always get something wrong by the director, no matter how good it is, because telling editors what they did wrong is how directors make money).

the more SFX-end ones like facial aging and face reshaper, or the talent-replacing ones like speech cloning/ADR, feel both too prescriptive for a director to dial in what they aesthetically want, and also not good enough for the final cut. so I struggle to find where they'd be actually useful in a workflow as opposed to being a trap, looking just fine enough at a glance to sneak into a final cut while looking poor when viewed by the audience.

likewise the focal adjustment and upscaling just feel gross. the kinds of things a good cinematographer can already do, and it'll be so so tempting to use tools instead of taking time to do it right in camera because it'll look good enough in the editing bay, but I feel like it'd really stand out as fake in the final cut unless you're targeting like, heavily compressed social ads. the less you use them the better they'll work, which isn't ideal for a marquee feature.

if the blemish remover really does respect continuity it'd be a nice-to-have, but it also feels like another trap to be careless/cheap on things like makeup or lighting at the shoot, at the expense of looking fake in post

I think that's the broader angle that bugs me the most. all of these tools are convenience tools for editors, but in the end they'll really be justifications for directors/producers/studios/agencies to cheap out and do shittier work faster on the shoot. a cheap, shitty shoot covered in AI bandaids is still going to hit an audience like a cheap, shitty shoot.