
ReAct agent
A plain-English definition, how it works, an example, and how it is used in enterprise AI.
What is a ReAct agent?
An AI agent that alternates between reasoning about what to do next and acting by calling a tool, using each result to decide the following step.
How it works
ReAct stands for reasoning and acting. Instead of planning everything up front, the agent works in a loop: think about the goal, choose a tool, call it, read the result, think again. The loop ends when the agent has enough to answer or finish the task.
The pattern makes agents more reliable, because each step is grounded in a real result rather than a guess. It also makes them easier to check, because the reasoning and tool calls can be shown step by step.
What a ReAct agent can do depends entirely on the tools it is given. In an enterprise, those tools should be defined, permission-checked and logged.
An example
Illustratively, an agent asked about a late payment first queries the invoice, sees it is unmatched, queries the delivery, finds it missing, and then recommends expediting the delivery instead of chasing the customer.
How NexusAI uses it
The NexusAI reasoning plane runs ReAct agents. On the practice enterprise, the copilot has 14 ReAct tools, and each answer shows a trail: which tools ran, the steps, whether the run was read-only and the tokens used.
Figures from the practice enterprise come from simulated data, not a customer.
Questions about ReAct agent
Why use ReAct instead of one big prompt?
Each step is based on a real tool result, so the agent corrects course as it learns, and every step can be shown and checked.
Can a ReAct agent use any tool it wants?
Only the tools it is given. In an enterprise, that list should be fixed, permission-checked and logged.
Get started
Build the intelligent enterprise
Turn enterprise systems into an intelligent operating layer that understands your business and executes operations autonomously.