Review of Multi-Agent Systems (MAS) in Education

Abstract

This systematic review synthesizes the evolution of
Multi-Agent Systems (MAS) in education from 2015 to 2025,
analyzing 523 peer-reviewed studies preliminarily. The
results reveal a significant surge in publications
post-2024 (n=112 in 2025), possibly driven by the
integration of Large Language Models (LLMs) and accessible
agentic frameworks. Higher education dominates the research
landscape (61.57%), partly due to its infrastructure and
flexible ethical constraints, while the K-12 and vocational
sectors remain underexplored. Pedagogically, MAS
applications prioritize subject-specific Knowledge Mastery
(n=128) and Learning Process Optimization (n=91), while
focusing less on broader 21st-century competencies like
collaboration and ethical development. This study
highlights a critical disconnect between technical
advancements and holistic educational needs, aiming to
provide a strategic roadmap for bridging the gap between
MAS innovation and pedagogical effectiveness in the
generative AI era.