AgentSchool: An LLM-Powered Multi-Agent Simulation for Education

Ye et al. (2026) introduce AgentSchool, an LLM-driven multi-agent simulator that models learning as state transition rather than prompted behavior. It couples cognitively growable student agents (weighted subject knowledge graphs, thinking-workflow pools, explicit misconceptions) with adaptive teacher agents that plan, scaffold, and reflect along the Zone Of Proximal Development, embedded in a configurable scenery generator and a multi-scale simulator. It produces more differentiated mastery and misconception traces than baseline simulators and generates plausible classroom social dynamics (peripheral participation, cliques, opinion-leader emergence).

The paper argues that validating educational AI is uniquely hard: interventions act on developing learners whose trajectories are irreversibly shaped, while real-world trials are slow, ethically constrained, and institutionally locked. LLM-based simulators offer a remedy, but many collapse learning into persona-conditioned role-play and can structurally penalize institutional novelty.

Architecture

Cognitively growable student agents: equipped with weighted subject knowledge graphs, thinking-workflow pools, and explicit misconceptions, so that their knowledge state changes as they "learn" (state transition, not just prompted persona).

Adaptive teacher agents: plan, scaffold, and reflect along the Zone of Proximal Development, adapting instruction to each simulated student.

Configurable scenery generator: situates instruction within both formal and informal learning fields.

Multi-scale simulator: decouples interaction scale, temporal granularity, and simulation duration.

Findings

Structured student agents produce more differentiated mastery and misconception traces than a baseline simulator — i.e., more realistic variation across learners.

Teacher-agent comparisons show backbone-dependent patterns consistent with ZPD-informed adaptation.

The simulator generates plausible social dynamics — peripheral participation, clique formation, aggressor-induced cohesion, and opinion-leader emergence — consistent with classroom social theories.

Implications

AgentSchool reframes student simulation as stateful learning rather than role-play, addressing the validity concerns raised elsewhere in the Simulating Students literature. It positions education as a testbed for long-horizon memory, multi-agent coordination, and institutional reasoning, while serving as a research instrument for validating educational AI and studying classroom dynamics.

Connected Concepts

Simulating Students

Agentic AI

Zone Of Proximal Development

Collaborative Learning

Connected Articles

Simulating Students Diverse Cognitive Levels 2025 — Embracing Imperfection: Simulating Diverse Cognitive Levels

Simulating Students LLM Review 2026 — Simulating Students with LLMs: A Review

Valid Student Simulation LLM 2026 — Towards Valid Student Simulation

LLM Student Simulation Misconception Faithfulness — Simulating Students or Sycophantic Problem Solving?

History Aware Student Simulation — History-Aware Profiles for Student Simulation

LLM Student Simulation Teacher Insights — Can LLMs Simulate Human Learners?

Citation

Ye, Y., Li, W., Wen, Z., Huang, Y., Hu, Y., Wei, Z., Wang, Y., Xie, X., Yang, H., Huang, Y., Li, R., Qian, H., Song, Y., Jiang, B., Li, B., Li, L., Zhang, B., Cai, P., Xu, X., Chen, S., Hu, X., He, L., Zhou, A., Qu, J., Shao, J., & Wang, X. (2026). AgentSchool: An LLM-powered multi-agent simulation for education. arXiv:2605.30144.