From Prompt Engineering to Dependable AI Job Engineering: A Progressive Expansion of the Engineering Object in AI Systems

This paper proposes a progressive expansion of the engineering object in AI systems: Prompt Engineering, Response Engineering, Persistent Job Engineering, and Dependable AI Job Engineering. Persistent Job Engineering treats a goal-oriented Job, rather than an individual response, as the unit of execution and introduces explicit Job identity, Runtime State, intent, decision history, Semantic Context Compression, suspension, and resumption. Dependable AI Job Engineering further addresses whether such a Job remains faithful to human intent, maintains consistent execution state and responsibility boundaries, recovers safely from failure, and reaches an assured completion condition. The paper relates this framework to yui-AI-OS, JDR, Runtime State, Semantic Context Compression, human selection gates, and JRPK, and argues that future AI Engineering should increasingly address the dependability of persistent human-AI work.