What should students not yet delegate to AI? Toward a theory of cognitive delegation boundaries in higher education

Generative artificial intelligence (AI) now performs tasks students previously had to perform themselves, from drafting arguments to summarizing literature. Existing guidance on this shift tends to ask whether AI use is disclosed or permitted, not which cognitive tasks a learner can safely hand over in the first place. This qualitative critical review uses thematic synthesis of 31 purposively identified sources spanning cognitive science, automation and trust theory, learning theory, self-determination theory, and empirical studies of generative AI in higher education to ask which academic tasks students should not yet delegate to AI at their current stage of development. The review identifies a recurring structural problem, the verification circularity problem: a student cannot reliably check an AI output on a skill they have not yet developed, because checking requires the competence being outsourced. Building on this problem and on classical automation-delegation theory, the review proposes the Cognitive Delegation Boundary Framework, which classifies a task, for a given learner, along two dimensions—skill-formation value and verification independence—yielding four zones: freely delegate, delegate with caution, delegate with reflection, and do not delegate yet. Zone membership is developmental: a task that a novice should not yet delegate may be appropriately delegated by an advanced learner. Current empirical evidence shows reduced effort, and some weaker performance, associated with heavy AI use; it has not yet tested lasting effects on learning. The framework is therefore offered as a conceptual proposal for empirical testing. The review argues that delegation guidance in higher education should complement disclosure and detection with this developmental, structural logic.