A.R.I.: A Personal AI Operating System for Longitudinal Human–AI Collaboration

This practitioner-grounded position and design proposal presents A.R.I., a personal AI operating system for longitudinal human–AI collaboration. Grounded in the author's experience using AI in marketing work, the design connects externally supplied model capabilities with user-owned context, revisable experience, current judgment and action, and work continuity through closure and later reuse. The first proposed empirical comparison asks whether explicitly reviewing the applicability of already available experience, together with the current problem's interpretation, criteria, and evidence, improves subsequent judgment beyond reminders, retrieval, structured experience use, additional computation, or post-draft review. Pre-decision refers to an externally controlled sequence in which experience review precedes task-response generation; it does not identify an internal neural attention event. Proposed outcomes include repeated correction burden, factual and scope errors, actual judgment and execution outcomes, and supervision and resource costs. No experimental results or demonstrated performance improvements are reported. The author welcomes feedback on the problem framing, relevant prior work, and the smallest informative comparison. The author, seon, is a marketing practitioner writing in a personal capacity. Substantive assistance from ChatGPT and Codex in conceptual refinement, literature discovery, drafting, translation, figures, and critical review is disclosed in the manuscript. The accompanying Korean manuscript is a translation of the same design proposal; the English manuscript is the primary version for citation.