Lesson 18: Reviewing AI Code and Challenging the Model

An AI coding agent reports, “Feature complete: all tests pass,” yet its diff adds an unrelated dependency, changes a public function, weakens error handling and deletes a validation check. The feature may work in ordinary use, but that result is not automatically acceptable. The central question is how strongly AI-generated code should be challenged before approval. Successful generation is evidence to examine, not proof that a change is suitable, safe or within scope.