FROM SCRIBE TO ARCHITECT A PRACTITIONER’S REFLECTION ON TEACHING ACADEMIC WRITING IN THE AGE OF LARGE LANGUAGE MODELS

Abstract

This paper begins from a classroom observation: students in an academic writing course who began, mid-semester, producing literature reviews of unusual fluency — prose that was carefully structured, appropriately hedged, and, on closer inspection, partly fabricated. The observation points to a structural problem. For decades, instructors have treated writing fluency as a serviceable proxy for intellectual engagement; generative AI has broken that proxy, not by improving student thinking but by making polished academic prose available without it. The central claim of this paper is that AI amplifies existing competence rather than compensating for its absence: students who already know how to read critically, formulate an argument, and evaluate a source use AI productively; those who do not are helped to sound as though they had been. Against this background, the paper proposes a concrete three-phase workflow —built around structured prompting, distributed research across multiple AI tools, and progressive conversation in a purpose-built AI environment — designed to teach the literature review as an act of design rather than transcription. The larger implication is pedagogical: if the course can no longer evaluate writing as a finished product, it must evaluate the decisions that produce writing. Teaching academic writing, in the age of large language models, means teaching judgment.