Lost in Translation: What the AI code debate keeps getting wrong