GPT does seem to stay out of the way and get things done. Only thing I notice is that the question tool/elicitation doesn't work that well any more so the thing doesn't stop to wait for input.

But I wonder if that's intentional because it can keep computing while you are answering, so long as your steer aligns well enough with the direction it wants to go. Better than letting a cache go cold and burning compute on bringing it all back up.