Abstract
The rapid emergence of generative AI presents libraries with both significant opportunity and genuine uncertainty. Unlike previous technology adoptions, AI raises complex questions about trust, accuracy, sustainability, and institutional capacity—often before clear best practices exist. Many libraries face external pressure to "do something with AI" without a framework for evaluating what that should mean or how to proceed responsibly.
This poster introduces a learning-first approach to AI exploration—an alternative to rushing toward production deployment. Drawing on a real-world case study informed by work at the University of Miami Libraries, the poster describes how AI initiatives can be intentionally designed as institutional learning rather than immediate service delivery. The approach prioritizes building shared understanding, technical literacy, and evidence-based decision-making across staff before committing to specific tools or services.
The poster walks through a proof-of-concept AI-powered search interface that served as a structured learning vehicle, giving staff hands-on experience with generative AI capabilities, prompt engineering, and knowledge retrieval. This experimentation surfaced critical lessons: the volatility of the AI ecosystem, model inconsistency, API instability, and substantial long-term maintenance burdens. These insights informed a strategic pivot toward managed AI platforms that prioritize accessibility, sustainability, and organizational comprehension.
Attendees will leave with a practical, transferable framework for designing AI initiatives that maximize learning, identify pivot points, and build institutional buy-in. The framework supports pragmatic, values-aligned decision-making and is intended to spark cross-institutional discussion about how libraries of all types can engage with AI responsibly.