AI Coding Agents
The class of AI tools that generate, review, refactor, and operate on code — either inline (IDE chat) or autonomously (agentic tasks with PRs). Examples: Claude Code, GitHub Copilot, OpenAI Codex, OpenCode, Cursor, Aider.Capability spectrum
Shared characteristics
- All benefit from explicit context/standards documents (see Agent Context Instructions)
- All introduce failure modes absent from human code (see AI-Specific Code Pitfalls)
- All require human approval as the final merge gate
- Output quality scales with the quality of instructions and context given
Agentic task safety model
Common pattern across platforms:- Agent works on isolated branch, not main
- Cannot merge without human approval
- All actions auditable (commits, PR history)
- CI/CD and branch protections apply normally
Use cases
- In-the-moment code improvement during development
- Large-scale systematic refactoring (see Ai Agent Technical Debt)
- Automated PR review (see AI Code Review)
- Security vulnerability detection and fix suggestion
Related
- Agent Context Instructions — how to align agent behavior to team standards
- AI Code Review — reviewing output from these agents
- AI-Specific Code Pitfalls — failure modes unique to AI-generated code