I cannot tell you how much time I have saved by stopping Claude and asking, "what are you doing?"

At least 50% of the time, Claude "realizes" it already has all the information but is doing something that's unnecessary for the current work, stop, and tell me the previous step has completed.

People complain about approval prompts etc and have Claude run in fully autonomous mode. Outside small bug fixes, I just never find that useful. It helps me immensely to see what commands Claude is running to understand where the work is going.