This paper documents an exploratory pedagogical intervention, financed through the ER2Digit European Digital Innovation Hub initiative, integrating Large Language Models into a Service-Design course at the University of Bologna. The study addresses the underexplored challenge of AI literacy development in completely "AI-naive" populations, investigating how fifth-year architecture students with zero prior exposure to generative tools appropriate LLM-based assistants within authentic public administration design projects. The intervention's original objective of advanced professional integration was radically reformulated into embedded alphabetization when initial assessment revealed students' complete unfamiliarity with conversational AI interfaces, privacy implications, and foundational mechanisms of probabilistic language generation. At the mid-point of intervention (week 6/12), preliminary findings reveal one unexpected pattern: intentional friction, where technical failures paradoxically generate deeper understanding of LLM behavior than seamless interactions. Two additional dynamics warrant post-intervention investigation: assessment blind spots in output evaluation, and ethical pragmatism emerging through situated practice. This work-in-progress contributes real-time documentation of AI appropriation under maximal readiness gaps—a context typically excluded from controlled studies but increasingly common in educational institutions.
C. Passerini (Sun,) studied this question.