AI Engineering: Design, Reproducibility, and Accountability in the Age of AI

Generative AI is becoming embedded in human intellectual work, including writing, design, programming, research, and decision support. Yet what is commonly called AI Engineering still focuses primarily on models, data, operations, and risk management, while lacking a unified framework for the design states that emerge through sustained human–AI collaboration and for their reproduction, migration, and succession. Starting from a comparison with conventional software engineering, this paper conceptualizes the body of information that materially governs AI behavior as an “AI Design State.” It argues that a central concern of AI engineering is not exact output identity, but the reproducibility of capability, intent, process, and accountability structures. It further distinguishes the portability of an AI Design State from its compatibility, and shows that transferring an AI work environment from one user to another requires not only state transfer but also Capability Transfer. Finally, Portability, Compatibility, Capability Transfer, Capability Preservation, and Capability Development are integrated under the concept of “Capability Continuity.” The paper proposes that a central task of AI engineering is to design uncertain AI as a reproducible and inheritable work system governed by human intent and accountability.