# Strands Agent Loop The agent loop is [[Strands Agents]]' core mechanism that turns an LLM into an autonomous agent. It is a recursive cycle: **input & context → reasoning (model) → tool selection → tool execution → feedback → response**. Each iteration appends results to the conversation history, so the agent's understanding compounds across turns until it produces a final answer. This is the concrete expression of the "model-driven" philosophy: the model, not hard-coded control flow, decides what to do next at each step. Understanding the loop is the prerequisite for everything else (tools, [[Strands Hooks|hooks]], [[Strands State|state]], interrupts), since those all hook into points of this cycle. ## Related - [[Strands Agents]] - [[Strands Hooks]] - [[Strands State]] - [[AI Agents]] - [[Agentic Engineering]]