Purpose of modifying state in `Human in the loop` section

Please correct me if I am wrong, but I do not feel the need to modify the state when users can simply ask a new question when interacting with an agent. How is it different? is it just to save tokens? or does it have impact on the performance of LLM or the LLM agent?