Kamus AI
Coding agent
An AI agent that reads, writes and runs code in a repository, beyond suggesting snippets.
A coding agent can change many files, run tests and make commits. That power needs guardrails: work on its own branch or worktree, grant the narrowest permissions for the task, and check the evidence (tests, build, diff) before trusting its report.
Also called: Coding agent, AI coding agent, code assistant
Related terms
Appears in 16 lessons
- What stays out of the model: secrets, production data, client code
- Code from AI is a draft, and you own it
- A written plan before code: plan, build, review
- Test first, and watch it fail for the right reason
- One task, one worktree, commits with explicit paths
- Reviewers who probe real data, not the summary
- Mini project: one small feature with an agent, from plan to hand-over
- What building with AI means
- From idea to spec
- Editors and CLI agents
- Troubleshooting production
- Project: release the team tool to a server
- Test first
- Session 1: machines ready and one idea
- Session 2: spec and written plan
- Session 3: build with tests and a separate reviewer