Transformative AI Agency: How Students Negotiate and Collaborate with Generative AI in Higher Education

As generative AI tools such as ChatGPT enter higher education, questions arise about how students can use them not merely instrumentally but as catalysts for collaborative and reflective learning. This study investigates how master’s students engage with ChatGPT in group-based academic tasks, specifically when working with complex course literature. Drawing on Vygotsky’s concept of double stimulation and Engeström and Sannino’s theory of transformative agency, we analyze how students collectively navigate AI-generated responses, challenge assumptions, and reframe understanding. The data stem from an exploratory case study in a Danish university course and include group discussions, ChatGPT logs, reflections, and focus group interviews. Findings show that ChatGPT mediates not only as a resource but as a mediating artefact that provokes resistance, negotiation, and generative questioning. Over time, students began to use technology not just for answers, but to explore and question ideas together. The study contributes to research on AI in education by highlighting the role of pedagogy in enabling transformative agency through Generative