Decoder. plain-English AI glossary

ReAct prompting

▲ Rising

A prompt structure where the model alternates between reasoning (thought), taking action (tool calls or steps), and observing results.

Think of it like

Planning → executing → reflecting, in a loop.

Example

"Thought: I need to find the GDP of France. Action: Search Google. Observation: France GDP is X. Thought: So compared to Germany... Action: Search Germany GDP. Observation: ... Thought: France is larger."

How it actually works

ReAct is foundational to agentic prompting. The structure forces the model to think, act, observe, and adjust. It works well with external tools (search, calculators, APIs). The observation loop creates feedback and prevents hallucination.

For product teams

Enables agent-like behavior: planning, tool use, correction.

For engineers

Implement with tool calling. Parse thought/action/observation. Loop until done.

Agents — Core to agent design.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome