A developer argument that AI agents make code quality irrelevant — analogous to not caring about compiler-emitted assembly — is examined and refuted on two grounds. First, LLMs are nondeterministic unlike compilers, so bugs cannot be systematically eliminated. Second, and more practically, agentic tools charge per token: messy, duplicated, poorly structured codebases force agents to consume far more context, directly inflating costs. SOLID principles, DRY, and separation of concerns turn out to reduce AI token usage just as they reduce human cognitive load, making good architecture an economic necessity rather than an aesthetic preference for AI-assisted teams.
