ReAct
ReAct is an agent technique that iteratively interleaves reasoning, actions, and observations so results can guide subsequent actions.
AI Agents in Action, Second Edition
ReAct is an agent reasoning technique that interleaves verbal reasoning with tool actions and observations in a feedback loop.
- How it is used
- Guides agents through multistep, tool-dependent tasks by using each result to choose the next action.
- How deep it goes
- Defines the closed loop, contrasts it with CoT, and demonstrates tool-using instructions without descending into model internals.
- Examples
- Uses backward-and-forward year tools and a trace-inspection exercise to make the action–observation cycle visible.
- What you can do after
- Build it from scratch as a tool-using agent that inspects results before proceeding, then verify calls in its trace.
Read how AI Agents in Action, Second Edition explains ReAct →
Domain-Specific Small Language Models
An agent framework that interleaves reasoning and tool use to solve tasks by iterating toward a solution.
- How it is used
- Deploys LLM agents that solve tasks by iteratively reasoning about problems and acting through external tools.
- How deep it goes
- Explains the iterative loop and demonstrates executable Python actions, without developing a formal theory of reasoning.
- Examples
- Uses flight search, seat selection, and ticket booking as chained actions in a travel task.
- What you can do after
- Apply it via smolagents by configuring an LLM, tools, and a CodeAgent for multistep task execution.
Read how Domain-Specific Small Language Models explains ReAct →
Build AI-Enhanced Web Apps
ReAct is an agentic reasoning technique in which a language model interleaves reasoning about a task with actions that gather information or affect an environment.
- How it is used
- Enables an agent to choose external tools and incorporate retrieved information into answers.
- How deep it goes
- Shows agent configuration and web-app integration, but does not detail the model’s internal reasoning mechanism.
- Examples
- Uses WikipediaQueryRun for technical questions and a chat application that displays search results.
- What you can do after
- Apply it via LangChain.js to configure a tool-using agent and integrate its responses into a streamed chat UI.
Build an AI Agent (From Scratch)
The ReAct pattern is an agent design in which reasoning and action alternate so the agent can incorporate tool results and adapt its subsequent behavior.
- How it is used
- Uses iterative LLM decisions and tool actions to solve multistep questions when the necessary steps are not known in advance.
- How deep it goes
- Develops the cycle into a tool-execution loop and named agent components, then identifies failures on complex tasks.
- Examples
- Traces web searches and a calculation for a runner-to-Moon estimate, plus a current-information search through a simple agent loop.
- What you can do after
- Build it from scratch as a tool-calling agent that iterates over results and returns a final or structured answer.
Building Reliable AI Systems
An agent framework that interleaves reasoning, actions, and observations to guide task execution.
- How it is used
- Guides agents through tool selection and changing multistep workflows by cycling between reasoning, action, and observation.
- How deep it goes
- Explains the decision cycle through a travel scenario and an agentic RAG implementation, without specifying a prompt format.
- Examples
- Uses a flight-and-hotel request and a travel assistant combining a weather API with a vector store.
- What you can do after
- Apply it via LangChain to coordinate retrieval and tools while exposing reasoning steps in a chat interface.
Prompt Engineering in Practice
ReACT is a prompting pattern for agentic workflows in which a model alternates between reasoning, actions, and observations.
- How it is used
- Structures agentic investigations that need external observations before reaching a supported conclusion.
- How deep it goes
- Works through a three-turn diagnosis and describes tool-runtime handoffs, but demonstrates the loop with manually supplied observations rather than code.
- Examples
- Traces stale-dashboard diagnosis through repeated hourly transformation failures and a schema error involving `user_segment`.
- What you can do after
- Practice/perform an observation-led investigation and stop when independent evidence supports a diagnosis and next action.
AI Agents and Applications
A ReAct agent is an agent architecture that interleaves reasoning with actions such as invoking tools and observing their results.
- How it is used
- Coordinates dynamic tool selection and sequencing when a user request cannot be mapped to fixed API calls.
- How deep it goes
- Explains the reasoning-action loop, tool contracts, orchestration, state inspection, retries, and tracing, but not model internals.
- Examples
- Demonstrates customer-order workflows, travel and weather lookup, and hotel-plus-B&B availability for Penzance.
- What you can do after
- Build it from scratch or with LangGraph’s prebuilt agent, then debug tool choices and execution through state inspection and traces.
Hugging Face in Action
ReAct is an agent technique in which a language model alternates between reasoning, taking actions such as tool calls, observing results, and continuing reasoning.
- How it is used
- Builds agents that reason through user queries and call suitable tools when their internal knowledge is insufficient.
- How deep it goes
- Explains the reasoning-action-observation loop and basic execution traces, but does not develop deeper internals or training procedures.
- Examples
- Demonstrates web search, weather lookup, factual questions, arithmetic, and message-history traces.
- What you can do after
- Build it from scratch with LangChain or LangGraph, configure tools, invoke queries, and inspect the resulting agent trace.
Build Python Web Apps with Streamlit
A ReAct agent is an AI agent pattern that interleaves reasoning about a task with actions performed through callable tools.
- How it is used
- Deploys agents that interact with existing systems instead of only retrieving facts.
- How deep it goes
- Explains the reasoning-action loop and tool interfaces, but defers the internal implementations of those tools.
- Examples
- Uses order tracking and cancellation, static user and order dictionaries, and four customer-service tool functions.
- What you can do after
- Build an agent that selects callable functions, incorporates their results, and completes customer order tasks.
Read how Build Python Web Apps with Streamlit explains ReAct →
LLMs in Production
ReAct is a prompting framework that interleaves a language model's reasoning with external actions and observations to solve tasks.
- How it is used
- Uses few-shot prompting to let language models solve tasks through successive reasoning, tool calls, and observations.
- How deep it goes
- Explains the reasoning-action-observation loop with a worked trace, then demonstrates agent configuration without detailing implementation internals.
- Examples
- Demonstrates swallow-speed research and an Olivia Wilde, Harry Styles, and exponent-calculation agent trace.
- What you can do after
- Apply it via LangChain by connecting an OpenAI model with search and math tools in a ReAct agent.
Building LLM Applications with DSPy
The ReAct pattern guides an agent to interleave reasoning with actions such as selecting and invoking tools.
- How it is used
- Builds agents that select Python-wrapped tools to answer queries, including chatbots that may combine ReAct with RAG.
- How deep it goes
- Explains iterative tool selection through trajectories of reasoning, tools, parameters, and execution outputs.
- Examples
- Uses tool-wrapped Python functions and a long-running chatbot loop as application patterns.
- What you can do after
- Apply it via DSPy to build an agent that selects tools and iterates from execution histories.
Read how Building LLM Applications with DSPy explains ReAct →
Related concepts
- memoryhas part · Memory is a storage resource or mechanism that retains data or computational state for access during computation or later use.
- tooluses · A tool is an instrument or software capability that enables a person or system to perform a task, obtain information, or interact with resources.
- AI agentkind of · An AI agent is a computational system that uses artificial intelligence to interpret information, select actions, and pursue goals through interaction with an environment.
- tool usageuses · Tool use is an AI system's invocation of external software functions, services, or other capabilities to obtain information or perform actions.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.