AI pair programming

A conversational coding workflow in which a developer and an AI assistant work through a change together, exchanging code, questions, explanations, tests, and feedback in short turns.

AI pair programming describes the assistant-like workflow popularized by tools such as GitHub Copilot. The developer can ask for a function, inspect a suggestion, run it, report a failure, and request a narrower revision. The assistant may autocomplete a line or draft a larger change, but the interaction remains a sequence of proposals and feedback rather than a separately running agent owning the whole task. The name borrows the social model of pair programming; the AI does not share legal or organizational responsibility for the result.

Builder example

The workflow is useful when the next decision benefits from immediate conversation: exploring an unfamiliar API, comparing implementations, explaining a failure, or drafting a focused test. Trust should come from evidence suited to the change. Reading a subtle authorization rule may matter more than reading generated boilerplate line by line, while executable tests can provide stronger evidence for routine transformations.

A developer asks Copilot or another assistant to draft a parser, then steps through edge cases and tests the output.

Use the conversation to expose assumptions, then test the parser's accepted fields and types, returned fields, and failure behavior.

Common confusion: AI pair programming is a description of the interaction pattern, not a rule that a person must read every generated line. A team can combine conversational assistance with automated checks, targeted review, or agent-owned patches. The required evidence depends on what the code can affect.