In agentic programming how do you handle follow up task

I had this issue when the agent is doing something and the user send another task to correct the first intent how do you handle it. Now I'm using debounce but that affects 95% of messages that do not require to be rectified