An agent is an LLM with tools and a loop

An agentic system lets the model plan, call tools, observe results, and decide what to do next. That loop turns a stateless API call into a stateful, multi-step workflow with real side effects.