Terminal coding agent

A terminal coding agent is an AI assistant you start from the command line, the text window developers use to run programs. You give it a task in plain words, and it reads the files in the project, changes them, runs the tests and reports back, all without a graphical editor open. Claude Code and Cursor are tools that offer a command line agent.

Working from the command line makes the agent easy to run in places where an editor is awkward, such as a remote server, an automated build or a scheduled job. It also fits teams who already do most of their work in the terminal. A non-technical owner will rarely use it directly, but it matters if developers or contractors build your systems and you want them to be able to automate routine changes. If nobody on the team touches code, you can ignore it.

• Ask whether the agent can run unattended, for example as part of an automated check, or only with someone watching.
• Check what it is allowed to do by default, and whether it asks permission before editing files or running commands.
• Test it on a small real task, such as fixing a failing test, and see whether it finishes without constant guidance.
• Find out how usage is metered, since long tasks can use a lot of the allowance.