Evaluating Agentic AI: The Multi-Layer Evaluation Model (MLE-A)

Abstract

The emergence of agentic artificial intelligence (AI) marks a shift from reactive educational technologies towards systems capable of autonomous multi-step planning, tool orchestration, and adaptive decision-making. As these systems contribute to structuring study pathways, generating feedback, and monitoring learner progression, elements of cognitive and metacognitive regulation become distributed between learners and AI agents. However, existing evaluation approaches in educational technology remain largely focused on usability, performance, and satisfaction, and are therefore insufficient to capture these dynamic and distributed forms of agency. This paper proposes the Multi-Layer Evaluation Model for Agentic AI (MLE-A), a conceptual framework for assessing the educational impact of agentic systems across five interacting dimensions: cognitive, metacognitive, affective, behavioural, and system-level governance. The model provides a structured basis for analysing human–AI collaboration and supporting future empirical evaluation.