Agents
ReAct pattern
ReAct alternates model reasoning, actions, and observations in a task loop.
How it works
Each observation updates the next decision instead of relying on a fixed initial plan. The loop needs a stopping condition and protection against repeated ineffective actions.
A concrete example
Illustrative: search once, inspect 3 results, calculate 2 values, then answer.
Do not confuse it with Chain of thought
ReAct couples reasoning to external actions; chain of thought alone may remain internal text.