ReAct

A prompting pattern where the model alternates between Reasoning (thinking out loud) and Acting (calling a tool). Each cycle of thought → action → observation tightens the agent's approach.

Example: Thought: I need today's weather. Action: call weather_api('Amsterdam'). Observation: 18°C, cloudy. Thought: Now I can answer.

In practice: ReAct was introduced in a 2022 paper and remains a foundational idea. Most modern agent frameworks implement some variant of it under the hood.

• is a pattern within Agentic LoopReAct is a specific reasoning pattern that describes how the agent thinks and acts inside each iteration of the agentic loop.
• builds on Chain-of-ThoughtThe 'Reason' step in ReAct is essentially chain-of-thought reasoning applied before each action.