This paper organizes how AI agents learn and improve into one simple map with four roads: A1, A2, T1, and T2.