As I say, those are the arguments that I actually hear people make. And most of the arguments I'm hearing people make are from people who are using LLM tech on a daily basis (not the agent stuff so much, although it sounds promising). I remain surprised you're encountering so many people quite so skeptical.
The only real dispute about the usefulness of the current technology I hear is the concern that this will leave us without devs who understand the code at a deep level.