Who Decides in AI-Mediated Learning? The Agency Allocation Framework

A team of researchers from Cornell University and KTH Royal Institute of Technology has published a significant paper titled 'Who Decides in AI-Mediated Learning? The Agency Allocation Framework' that addresses a critical gap in educational technology. As AI systems like tutoring platforms and adaptive learning tools increasingly shape educational experiences, the researchers argue that learner agency—who actually controls decision-making—has become both more consequential and harder to conceptualize at scale. Their framework reframes agency as the allocation of decision authority across four key stakeholders: learners, educators, institutions, and the AI systems themselves.

The Agency Allocation Framework (AAF) provides systematic tools for analyzing how decisions are distributed, how choices are architected, what evidence supports them, and over what time horizons consequences unfold. Drawing on Learning@Scale literature and practical examples from tutoring systems, the researchers identify four recurring challenges: conceptual ambiguity about what agency means, over-reliance on behavioral proxies like engagement metrics, inherent trade-offs between efficiency and learner control, and the fundamental redistribution of agency that occurs when AI mediates learning processes.

Rather than advocating for more or less automation, the AAF supports nuanced analysis of when AI should scaffold learners' capacity to act versus when it substitutes for their decision-making. The framework makes decision authority explicit, providing researchers and designers with analytic tools for studying, comparing, and evaluating agency-preserving learning systems. This work, accepted to the ACM Conference on Learning @ Scale (L@S '26), comes at a crucial time as educational institutions worldwide grapple with implementing AI tools while maintaining meaningful learner autonomy.