The second essay in a series on AI trust and practice theory. Building on the first essay (AI Trust and Situated Awareness: A Practice Theory Reframe), this paper turns on a problem the framework left implicit: if a practitioner's situated awareness about a practice is captured in a transmissible form, who carries the about-the-user knowing that holds across whichever practice the LLM is currently enacting? The answer is apprenticeship as a layer separate from any individual practice — a standing arrangement between the harness LLM and the user within which discrete practices are reached for. The apprenticeship is shaped as an MCP server that projects the engagement and the active practice together; the LLM enacts the projected practice and is apprenticed simultaneously in the practice (its teleo-affective, understanding, rules, affordances, and materials) and about the user (preferences, history, current concerns). The essay then sketches an autonomic loop that maintains the apprenticeship through judge-and-smoother practices acting on the enactment history — and notices that those practitioners are themselves practices, opening onto a strange loop in which the apparatus improves itself through the same machinery it provides. (The essay was written before two later refinements in the series — "meaning layer" was sharpened to "situated awareness" and "practice recipe" to "practice bundle"; see the editor's note at the top of the paper.) The third essay implements the apparatus in twelve steps; the fourth demonstrates it on a concrete calendar-stewardship case.
Monyet Batu (Wed,) studied this question.