A Scoping Review of Large Language Model-Based Pedagogical Agents

A comprehensive scoping review of 52 studies examines Large Language Model-based pedagogical agents across educational contexts from November 2022 to January 2025. The research identifies four key design dimensions (interaction approach, domain scope, role complexity, system integration) and emerging trends including multi-agent systems, virtual student simulation, and integration with immersive technologies, while flagging critical research gaps around privacy, accuracy, and student autonomy.

This scoping review documents the rapid maturation of LLM-based educational agents, a field that has evolved dramatically since the public release of advanced language models in late 2022. The analysis of 52 peer-reviewed studies reveals that researchers are moving beyond simple chatbot implementations toward sophisticated multi-agent architectures that simulate realistic learning environments. The identified design dimensions provide practitioners with a conceptual framework for building and evaluating these systems across diverse educational contexts.

The research represents a natural evolution in pedagogical technology. Traditional intelligent tutoring systems and conversational agents have long aimed to personalize instruction, but LLMs introduce capabilities in reasoning, knowledge synthesis, and contextual understanding that exceed previous generations. The breadth of applications—spanning K-12, higher education, and informal learning—demonstrates rapid adoption across educational sectors and subject domains.

For the EdTech industry and educational software developers, this review signals significant commercial and research opportunity. The emergence of multi-agent systems and integration with learning analytics suggests that pedagogical agents are becoming infrastructure components within broader educational platforms rather than standalone tools. The integration with immersive technologies (AR/VR) points toward hybrid learning experiences that combine conversational AI with spatial computing.

Future development hinges on addressing critical gaps around privacy protections, ensuring factual accuracy in agent responses, and preserving student agency in learning decisions. Researchers and developers should prioritize evaluation methodologies, ethical frameworks, and transparent disclosure of model limitations. The field's sustainability depends on moving beyond proof-of-concept implementations toward rigorously validated systems that demonstrate measurable learning outcomes.

• →LLM-based pedagogical agents span K-12 through higher education with four distinct design dimensions: interaction approach, domain scope, role complexity, and system integration.
• →Multi-agent systems simulating naturalistic learning environments represent an emerging trend combining conversational AI with learning analytics and immersive technologies.
• →Critical research gaps exist regarding privacy protections, factual accuracy of agent outputs, and preservation of student autonomy in learning processes.
• →The field has evolved rapidly from November 2022 to January 2025, with 52 documented studies indicating growing adoption across diverse educational contexts.
• →Future development requires rigorous evaluation methodologies, ethical frameworks, and transparent disclosure of model limitations rather than continued proof-of-concept implementations.