When introducing automated programming tools, one should pilot them on low-risk, high-repetition tasks first, and gradually build empirical judgment about the boundaries of their capabilities.