AI should only run as fast as we can catch up

Through conversations with two friends using AI differently, a pattern emerges about AI's reliability depending on verification speed. Eric, a product manager, generated impressive-looking web apps with AI but lacked technical knowledge to verify their production readiness, while Daniel, a senior engineer, successfully shipped AI-generated code because he could quickly validate the outputs. This highlights that AI works best when verification effort is much lower than creation effort, creating "verification debt" where organizations risk deploying unvalidated AI work.