I see this as the gap between an general-purpose agent and a coding agent. A coding agent can imagine something to be true, test it, discover that it's wrong, and recover.
But if you go beyond what can be tested easily, asking the agent to do real work rather than writing a patch, imagining things to be true is a problem.
But if you go beyond what can be tested easily, asking the agent to do real work rather than writing a patch, imagining things to be true is a problem.