Comment by Terr_

I think it's more like "LLMs agents are especially dangerous for beginners in $X."

Those are the users who won't have (as much) necessary knowledge, unused mental-bandwidth, or "instinctive taste" to recognize things that need validation and then fact-check them.