Large language models are being used as autonomous agents that interact with environments, plan, and recover from mistakes, but current post-training methods fail to fully utilize the feedback provided by these environments. This is because prevailing methods, such as reinforcement learning with verifiable rewards, focus primarily on optimizing final success signals, rather than leveraging the rich feedback available. Researchers argue that this oversight leaves significant potential for improvement untapped, as environment feedback can provide valuable insights for agent development. The proposed approach, internalizing agency from reflective experience, aims to address this limitation by incorporating reflective experience into the learning process1. This could have significant implications for the development of more effective and adaptive autonomous agents. So what matters to practitioners is that this new approach could lead to more efficient and resilient AI systems, capable of learning from their interactions with complex environments.