Three Observations from Pair Programming with AI
Certainty: First-hand experience, still a small sample
I’ve been leaning heavily on AI to write code lately. Three quick notes to start:
- The roles are flipped. In classic pair programming the novice drives and the veteran navigates; when I pair with an AI, the AI drives and I navigate—my value shifts from “being able to write it” to “being able to judge whether it’s written correctly.”
- Review has become the bottleneck. The AI produces code far faster than I can read it line by line, so “how to verify quickly” (tests, builds, examples) is worth optimizing more than “how to generate quickly.”
- Prompt quality sets the floor; acceptance criteria set the ceiling. More on this in Writing Prompts as Specs.
Belongs to: AI Coding Workflow