Human agency is fundamental to meaningful learning and self-regulated learning, yet it is increasingly challenged and reshaped by the rapid integration of (generative) AI into education. This talk examines how AI can either encourage cognitive offloading or support cognitive empowerment, depending on how human-AI interactions are designed and enacted. Particular attention is given to reading and writing–core practices through which learners develop comprehension, articulate ideas, construct knowledge, and form their identities as readers, writers, and thinkers. Drawing on perspectives from the learning sciences, human-computer interaction, literacy research, and responsible AI, the talk conceptualizes four interrelated dimensions of agency in AI-mediated learning: human oversight, human-AI complementarity, AI competencies, and relational emergence. It explores tensions surrounding autonomy, transparency, trust, and cognitive offloading, including the risks of outsourcing interpretation, reasoning, and text production to AI. Through emerging research and design examples involving AI-supported reading and writing, the keynote considers how AI can scaffold comprehension, reflection, goal setting, feedback, and revision while preserving learners’ intellectual ownership and teachers’ professional judgement.
