There are good use cases and bad ones.
Nothing wrong with an OCR that uses a little machine learning and statistical error correction to accurately read text on a page so it knows when something is a O and when something is a 0. Or a machine learning image recognition system that can point out critical elements in a scene (without mistaking a bag of chips for a gun for example).
Big difference between those assistive tools and things like these so-called "agent" features being added to big-name applications that claim to try and do your work for you. Or offer a summary that winds up being the polar opposite of the text it was summarising.
There is a class of user out there, who thinks the latter is the bees knees and thinks "AI" should be in everything, and to hell with irritating things like copyright law or responsible consumption of resources. Frequently these are people that do not have any physical impairments to using a computer normally and would not see value in something that can read text off a page for them.
They're the lazy ones.