it would be nice if an agent reply could ask for a follow-up agent reply if it's not done thinking about it (basic, readable reasoning0