An AI agent is an LLM running in a loop that can call tools (APIs, databases, browsers, terminals) and act on the results, iterating until a task is complete. Where a chat assistant produces one response, an agent can read files, run code, browse the web, and edit a system without a human in the loop on every step.
Agent quality in 2026 is improving but unreliable for sustained autonomous work; most production agents run with bounded scopes, careful guardrails, and human review checkpoints. Cursor, Devin, Lindy, and Operator are leading agent platforms; LangChain and AutoGen are the popular frameworks.