A coding agent wraps a large language model in a loop with tools: it can search and read files, execute shell commands, apply edits, and observe the results of what it just did. That feedback loop is what separates it from autocomplete or chat assistants, because the agent can run the tests it wrote, see the failure, and fix its own mistake without a human relaying the error.
Examples include Claude Code, OpenAI Codex, and Cursor’s agent mode. In practice the quality of an agent’s output depends less on the underlying model and more on the harness around it: the instructions, verification checks, and context it is given to work with.