From Prompts to Natural-Language Programs: Job Descriptors as Executable Specifications in yui-AI-OS — A Case Study of Job Descriptor No.002

This paper proposes the concept of a natural-language program executed by AI, using Job Descriptor No.002 in yui-AI-OS as a case study. In conventional use, a prompt is often treated as a request for an answer or a short task. By contrast, a Job Descriptor specifies an executable process: it defines a purpose, procedures, conditions, repeated interaction, user input, outputs, logs, feedback, and improvement cycles. The central claim of this paper is that a Job Descriptor is not merely a prompt. It is an executable specification written in natural language and interpreted by AI as a runtime engine. Job Descriptor No.002, a Human-Guided Daily Thinking and Reflection Job, demonstrates this claim through a daily dialogue process in which the AI asks questions, adapts to the user, records a daily intellectual history, and improves subsequent runs. The paper analyzes the descriptor as a user-level program, distinguishes it from ordinary prompting, and discusses the possibility that natural-language Job Descriptors may become a major form of end-user programming in the AI era.