# 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