A lot of the sci-fi versions of AI are about systems that slowly stop asking permission. In real products, I keep noticing the smaller version of that: the assistant that fills in blanks, nudges choices, and starts acting like it knows the user better than the user does.
That can feel helpful right up until it starts steering the work instead of supporting it. Has anyone run into a product where the AI felt subtly too confident, and people only noticed after it had shaped the workflow?
