Release Engineering for AI Features

AI features weaken the assumption that tested code equals deployed behavior because model providers, prompts, retrieval indexes, and safety rules can change without a conventional deployment event. Release engineering must manage prompts, models, retrieval data, and tool contracts as release inputs, using canary rollouts with behavioral health signals and consequence-based traffic segmentation.