It's actually less fatiguing than interacting with humans, who never admit they're wrong or agree to look at things from a fresh perspective.

Yes, very familiar. I think it might be ok though, as it's similar to a web search where you maybe need to refine it a few times. Just assume that the first answer may not be fully correct or complete, until the AI has more context. It would be a nice change if it would ask the follow-up questions automatically.

Fatigued, but not from your chosen examples.

With agentic coding, it is like when I play a 4X game such as Civilization: "just one more turn".

I don’t even read the conversational garbage. Only care about the outputs I demanded it give me.