pair-programming

AI pair programming lets an autonomous agent act as the primary implementer under human supervision. You initiate a session using a command like `/pair #123`, and the agent manages the workflow through planning, implementation, and automated review loops before reaching a human checkpoint.

An automated review loop is a workflow phase where the AI agent runs automated code checks before requesting human oversight. It ensures code quality by having the agent iteratively review and fix issues, so the human supervisor only handles final approval and critical architectural decisions.

AI coding agents manage parallel development by leveraging sub-agent orchestration and git worktrees. This allows the agent to isolate feature branches, implement code concurrently, and maintain structured execution across distinct phases like Trigger, Planning, and Merge.

Yes, autonomous code generation can integrate with issue-driven delivery to ensure structured execution. The AI agent reads the issue, acts as the primary implementer, and automates the development workflow until a human checkpoint is reached for final review.

You should avoid AI-driven pair programming when a feature lacks a well-defined issue or requires complex, highly ambiguous architectural decisions upfront. This approach relies on human oversight at key checkpoints, so it is unsuitable for unstructured tasks needing continuous manual exploration.

Agent-led pair programming integrates with persona-switching skills to ensure proper attribution during autonomous implementation. This combined workflow allows the agent to adopt specific developer profiles while managing code generation and review loops under human supervision.