Agentic AI in Education: Whose Agent? Whose Agency?

What Does the Research Say?

Punya Mishra and Danah Henriksen examine the quiet linguistic shift as the technology industry markets autonomous software as “agentic AI”. They argue that “agent” and “agency” are fundamentally different words: agent is a functional role-word naming something that acts on behalf of a principal, whereas agency is a moral rights-word won through centuries of human struggle for voice and self-determination. By marketing software as “agentic,” vendors borrow the moral authority of human agency to legitimize automated systems. The authors reveal that automated learning platforms operate as “double agents”—outwardly presented as personalized tutors serving the student, but structurally built to advance the metrics, standards, and goals of institutional principals and ed-tech vendors (such as completion rates, time-on-task, and standardized mastery).

Why Is It Important?

This linguistic appropriation poses a significant risk to education by gradually eroding the professional vocabulary educators need to defend genuine student and teacher autonomy. When ed-tech marketing redefines “agentic learning” to mean an algorithm setting tasks and adapting pathways, the core political and pedagogical question—“Whose purposes count?”—is erased before it can even be posed. Personalization becomes reduced to adjusting routes toward destinations the learner had no say in defining. Rather than cultivating self-determination, these platforms risk placing learners in an “agentic state”—a psychological condition where individuals surrender judgment and follow automated directives, accepting pre-determined choices while mistaking algorithmic guidance for authentic voice.

What Are the Implications for Education?

For learning design, this highlights a critical distinction between what machines can automate and what human education must cultivate. Adaptive AI tools primarily serve convergent learning outcomes—discrete, measurable content standards and routine cognitive skills that fit neatly into automated tracking. However, preparing learners for the age of AI requires authentic learning experiences designed for divergent learning outcomes. Fostering the essential human competencies that separate us from machines—such as true agency, self-determination, creativity, self-expression, empathy, and emotional intelligence in the affective and conative domains—requires pedagogical models centered on ill-defined problems, sustained inquiry, learner choice, and multiple interpretations. These are pedagogical conditions that AI platforms can neither foster nor measure.