I'll disagree, I think AI meaningfully changes the definition of a bad coworker because its main interface emulates talking to another human being closely enough that some people value its conversational input equally or more highly than that of actual human beings.

I think that this is exacerbated by many LLMs defaulting to placating, complementing, and otherwise reinforcing biases in their users, and it will often produce just enough evidence to make its (and its user's) position seem compelling, while that evidence is often either tangential, orthogonal, or straight-up hallucinated.

Humans aren't perfect vessels for information, but working on a reasonably-qualified, healthy team, a degree of respectful professional disagreement has always been the norm, and that covers both cases where the disagreement comes from a genuine matter of opinion, as well as cases where one colleague cross-checks anothers' assumptions and finds some of them to be wrong. That's the process.