AI glossary
Coding agent
An AI agent that works on a codebase: it reads files, edits code, runs commands and tests, and iterates until a task is done.
A coding agent goes beyond autocomplete. You describe a change (“add a login page”, “fix this failing test”) and it explores the project, writes the code, runs it and fixes its own errors. Claude Code, OpenAI’s Codex, Cursor’s agents and Muse Code are examples.
Code suits agents well because results can be checked automatically. Still, review what they produce: an agent can make a test pass in the wrong way or change more than you asked.
Example: You ask it to “add a button to export to CSV on the orders page”. The agent finds that page, writes the code, adds a test, runs it, fixes an import error and leaves the change ready for you to review.
In practice
- Work on a branch or a copy, never directly on what is in production.
- Ask for small, specific changes and review the diff before accepting them.
- Explain the project’s rules (how tests run, the style, what not to touch) in an instructions file it always reads, such as
CLAUDE.mdin Claude Code orAGENTS.mdin Codex.
More in what an AI agent is and in the Cursor Projects story.
