Action-Observation Loop
Domain: Prompt Engineering
🎯 Motivation
In ReAct, the model takes an action, observes the result, then reasons about what to do next. This loop continues until the task is done.
📋 Concrete Example
Loop: 1) Think: ‘Need stock price.’ 2) Act: get_stock(AAPL). 3) Observe: ‘$185.50’. 4) Think: ‘Price is up 2%. Now summarize.’ 5) Act: respond. The loop drives the agent forward.
🔗 Analogy
A scientist running experiments - hypothesize (think), experiment (act), observe results (observe), hypothesize again. The action-observation loop is the scientific method applied to AI agents.