A repeating cycle of **Observe** (gather information), **Think** (decide what to do), and **Act** (execute)—the core pattern behind many autonomous AI agents.
ReAct stands for Reasoning and Acting. In each cycle, the system:
- Observes — reads files, tool outputs, errors, or other signals.
- Thinks — chooses the next step toward the goal.
- Acts — runs a tool, edits code, calls an API, etc.
The loop continues until the goal is met or the system stops. Debugging agent behavior often means asking which phase failed: wrong observations, flawed reasoning, or a bad action.