AI Trust: Practice Theory — The Apprenticeship and a Strange Loop

The second in a series of essays on AI trust and practice theory. Building on the framework established in "AI Trust and the Meaning Layer: A Practice Theory Reframe", this paper describes an apprenticeship process — an MCP server through which any harness and their LLMs can be apprenticed to both the user and the user's practice recipes. It then describes how the recipes are maintained over time through a two-loop autonomic architecture (Judge and Smoother) that observes its own operation, drawing parallels with IBM's MAPE-K work in autonomic computing and with Hofstadter's strange loops. The result is a self-maintaining apprenticeship that survives across harnesses and is owned by the user rather than the vendor.