ModelRefs / ReAct — Agent Pattern
ReAct — Agent Pattern
Interleave reasoning with tool calls in a Thought→Action→Observation loop until the task completes.
Overview
ReAct alternates between reasoning steps and tool invocations. The model writes a Thought, picks an Action (tool call), reads the Observation, and repeats until it emits a Final answer.
When to use it: You need a simple, debuggable autonomous agent that can use tools and reason about results.
Pattern details
| Pattern class | planning |
|---|---|
| Difficulty | beginner |
| Autonomy | guided |
| Also known as | thought-action-observation, reasoning and acting |
| Last reviewed | 2026-06-07 |
Known failure modes
- Infinite loop — Agent keeps invoking tools without progress. Mitigation: Cap max steps; detect repeated tool calls.
- Hallucinated tool call — Model invents a tool that does not exist. Mitigation: Validate tool name and arguments before execution; return strict error.
When not to use it
- Exposing destructive tools without approval.
- Running ReAct on a model with weak tool calling.
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to ReAct — Agent Pattern.
Frequently asked questions
When should I use the ReAct agent pattern?
You need a simple, debuggable autonomous agent that can use tools and reason about results.
What are common failure modes of ReAct?
Infinite loop • Hallucinated tool call
Is ReAct production-ready?
Yes when paired with the safety controls and observability hooks documented on the pattern page.