“The dominant paradigm for training AI agents has shifted to reinforcement learning, which requires building complex "RL environments."”