This post argues that the best use of AI in software development isn't in code generation, but in facilitating understanding and exploration of changes in code. It emphasizes focusing on specific questions and maintaining a clear mental model, positioning code as a secondary product of this process.