Coding agent
What is Coding agent?
An AI system that carries out software tasks with limited supervision: it reads the codebase, plans an approach, edits files, runs tests and fixes what breaks. Purpose-built coding-agent tools work this way. It is the actor in agentic coding, and it performs best on well-scoped tasks with clear acceptance criteria.
Why it matters
An agent is only as safe as the environment you put it in. To do its job it needs real access: the repository, a way to run commands, a test harness, sometimes credentials. That access is exactly what makes an unscoped agent dangerous, because a system that can edit files and run commands can also delete the wrong ones or push somewhere it should never touch. The operational question is therefore not how clever it is but what it can reach, and who checks before its work lands. Teams that run these well treat the agent like a new contractor with broad tools and no context: a sandbox to work in, permissions scoped to the task, and a human gate before anything merges.
In practice
A team gives an agent its own branch and a disposable environment, read access to the codebase, and no keys to production. It works the task, opens a pull request, and a person reviews before merge. The setup is deliberately boring. It is what lets you use the capability without betting the repository on the agent never making a bad call.
