Who Holds the Agency? Meta-Design in the Age of Generative AI

The increasing adoption of Large Language Models (LLMs) in healthcare opens new opportunities for supporting caregivers and patients, but also raises important challenges related to reliability, transparency, and human control. In this work, we explore how a meta-design perspective can guide the development of AI-assisted End-User Development (EUD) environments that enable non-technical users to define and evolve system behaviors through natural language interaction. We present a conversational system that allows caregivers to create and modify therapy plans for patients through interaction with an LLM-based assistant. The system combines the semantic reasoning capabilities of LLMs with deterministic validation mechanisms and structured knowledge management to ensure safety, consistency, and traceability. Based on this design experience, the paper outlines a set of design implications for AI-assisted EUD systems, highlighting how meta-design principles can help balance the flexibility of generative AI with the need to preserve transparency, accountability, and meaningful user agency in safety-critical domains.