AI pair programming
AI pair programming works alongside an AI coding assistant (Claude, Copilot, Cursor, Continue) as a continuous collaborator on coding tasks, suggesting completions, generating tests, explaining unfamiliar code, drafting refactors. The discipline differs from one-shot AI prompting: the model stays loaded with project context across many turns within a single session.
Adoption data through 2025 shows AI pair programming is now the dominant assistance pattern for individual engineers, with measurable productivity effects strongest for routine tasks (boilerplate, test writing, syntactic transformations) and weakest for high-judgement work (architecture decisions, performance debugging). The practice changes the skill mix expected of engineers: less time on typing, more time on prompting, reviewing, and integrating AI output. Anti-patterns observed in deployed teams: accepting AI suggestions without reading them (introduces subtle bugs that compound); using AI to generate tests for AI-generated code (tests pass against the bug); skipping code review for AI-authored changes. Healthy pattern: AI generates, engineer reviews, human reviewer approves. The AI is a junior pair, not a senior one.