New study warns agentic AI may undermine student learning and agency

A new research paper accepted at the AIED 2026 Festival of Learning workshop tackles a critical question: as AI in education moves from simple chatbots to proactive, goal-directed agents, are we sacrificing genuine learning for convenience? The study, led by Steve Woollaston and colleagues from Kyoto University, reviews six core pedagogical principles—prior knowledge activation, collaborative learning, problem-based learning, formative assessment, scaffolding, and metacognition—through the lens of agentic AI. The authors argue that while these proactive systems can personalize learning, they risk automating away the very cognitive effort that drives deep understanding.

The paper proposes four design recommendations to maintain the balance: intentional friction (adding deliberate obstacles to promote thinking), dynamic scaffolding (adjusting support in real-time based on learner needs), human-in-the-loop oversight (keeping educators involved), and considered AI utilization (using AI only when it genuinely enhances learning). The research directly addresses the tension between automation and agency, urging developers to prioritize learning outcomes over seamlessness. This work arrives as schools and EdTech platforms rapidly adopt agentic AI for tutoring, grading, and curriculum planning—making the paper’s warnings especially timely for educators, technologists, and policymakers alike.