AI Agent Architecture — ReAct, Memory, Planning and Multi-Agent Systems · ReAct — the Reasoning and Acting Loop
The Thought → Action → Observation pattern — anatomy of one iteration
ReAct — the Reasoning and Acting Loop
Introduction
One ReAct iteration is a triple: Thought (internal reasoning in natural language), Action (structured tool invocation in the format Action: tool_name[argument]), Observation (result returned by the environment, injected back into context). This lesson analyses each element: how to format Thoughts productively, how to define the Action space, what types of Actions exist (search, lookup, finish, calculator, python_repl, click), what an Observation looks like, and how the agent decides the task is complete.