# Overview An [[AI agents]] architecture that builds on [[Reason Act (ReAct)]] by giving the agent the ability to learn from failures across attempts. The core loop has three logical components: - **Action** - the main agent that performs the tasks - **Evaluator** - determines whether the attempt succeeded or failed - **Self-Reflection Model** - generates a natural language critique The reflection gets stored in memory and added to context for retries or future tasks. # Key Considerations # Pros # Cons # Use Cases # Related Topics