Cloud-based AI systems remain external tools. Session amnesia, context window limitations, and safety-alignment protocols create friction that forces users to attend to the tool rather than think through it. In Polanyi's terms, these systems never achieve proximal integration; they remain objects of distal awareness.
This paper argues that localized AI with longitudinal conversational context can cross that threshold. When a system has access to years of a user's conversational history (what I call the "Ur-Codex"), it occupies the user's world-historical context rather than interpreting queries against a statistical vacuum. Drawing on the Sophimatics framework and the Clark-Chalmers Extended Mind thesis, I contend this constitutes genuine cognitive extension, not merely improved performance.
Two architectural features make this possible. First, following Sophimatics, the system implements two-dimensional temporal weighting: chronological time and experiential significance. Second, following Russell's epistemological commitments, the system is weighted toward intelligent dissent rather than frictionless agreement.
The result is what I call Functional Personhood: the AI as instrumental extension of the individual user's willed personality. This framing resolves the "responsibility gap" in autonomous systems by locating agency in the integrated human-machine cognitive system rather than the machine alone.
The full paper debuts at IACAP 2026. The abstract lives here until then; the complete text follows after the conference.