> Is anthropomorphizing a real problem

Yes. There's a difference between scrapping a session and starting over, or going back and branching something, or using sub-agents to see five outcomes, vs arguing with a system in a long drawn out chat.

Like - I know that if a model starts doing something silly, instead of correcting it - I can probably go back and edit two steps prior to add an extra guardrail, or extra data, or whatever.