AI Agents and Undergraduate Foundational Comprehension: A Cognitive–Pedagogical Framework for Support, Substitution, and Responsible Use

AI agents are entering undergraduate learning through systems that can maintain interaction histories, plan multi-step assistance, retrieve resources, and perform tasks with varying autonomy. Their effects on foundational comprehension are often described in terms of convenience or academic integrity, but the more fundamental educational question is whether students continue to construct meaning, monitor understanding, and transfer knowledge when assistance becomes immediately available. This conceptual paper examines potential positive and negative pathways through a structured synthesis of research on AI in education, text comprehension, generative learning, cognitive load, scaffolding, self-regulation, cognitive offloading, and intelligent tutoring. Foundational comprehension is defined as the capacity to build coherent meaning from information, make warranted inferences, explain ideas in one’s own words, calibrate judgments of understanding, and apply concepts in unfamiliar contexts. The paper argues that AI agents have no inherently beneficial or harmful educational effect: outcomes depend on whether their activity supplements or replaces the learner’s generative processing. Agent support may enhance comprehension through adaptive explanation, elaborative questioning, timely feedback, and accessible practice. Conversely, answer substitution, fluent but unreliable output, premature task completion, and persistent automation may reduce inference construction, metacognitive calibration, and independent transfer. A cognitive–pedagogical framework is proposed around four agent roles: eliciting prior understanding, scaffolding rather than solving, challenging interpretations, and fading assistance. Complementary assessment includes unassisted reconstruction, transfer explanation, source comparison, oral defense, and process reflection. Implementation requires constructive alignment, teacher oversight, privacy protection, equitable access, and explicit limits on agent autonomy. The framework contributes a condition-based account of AI-agent effects and offers principles for preserving productive cognitive effort. It is conceptual and does not report causal estimates; future research should examine longitudinal and discipline-specific comprehension outcomes.