AI Trust and Situated Awareness: A Practice Theory Reframe

For an LLM to be trusted to act as expected, it needs more than access to APIs — it needs the situated awareness a practitioner brings to the work: knowing where it is, what the goal is, what is at hand, and what would be a legitimate next step. The challenge has been how to express that knowing in a transmissible form. This essay draws on the human social sciences for the answer, looking to practice theory (Schatzki, Reckwitz, Shove) to introduce the practice bundle — a text-and-engagement form of apprenticeship through which an LLM is cued into a practice rather than taught it. (The essay was originally framed in terms of "the meaning layer" and a "practice recipe"; subsequent essays in the series sharpened both terms — see the editor's note inside the paper.) Unlike a carpenter who serves many clients, an LLM instance is enacted as a unique participant for one user, putting every user in the position of having a private practitioner. The essay closes by sketching an autonomic loop in which the bundle maintains itself through judge-and-smoother practices acting on the enactment history — a thesis the second essay develops, the third essay implements in twelve steps, and the fourth essay demonstrates on a concrete calendar-stewardship case.