ReAct
A prompting pattern that interleaves reasoning traces with actions.
ReAct (Reason + Act) prompts the model to alternate between a 'Thought' (free-form reasoning) and an 'Action' (a tool call), then feeds back an 'Observation'. Introduced by Yao et al. in 2022, it was the first widely adopted agent loop and remains the baseline most benchmarks (τ-bench, AgentBench, OSWorld) evaluate against. Native function calling has largely replaced the text-format parsing, but the interleaving structure survives.