XNews

Bob Martin: 'I Don't Read AI-Generated Code' — Uses Tests, Metrics, and Agent Pipelines

Bob Martin, author of "Clean Code," stated he does not read AI-generated code because human reading is too slow. Instead, he sets up layers of safeguards: unit tests, Gherkin tests, QA, code quality metrics, mutation testing, and coverage. He uses a four-agent pipeline for requirements specification, coding, refactoring, and architecture review. He monitors metrics like cyclomatic complexity, dependency structure, and module size to infer quality. He argues that when AI generates code faster than humans can read, code review shifts to writing tests, setting constraints, and defining metrics. He offers courses "AI Agents for Clean Code" (O'Reilly) and "Clean AI: Agentic Discipline" (Clean Coders platform).

Source posts