Multi-Agent Empowered AI-Generated Digital Educational Resources Development: Model Construction and Implementation Safeguards

Artificial Intelligence Generated Digital Educational Resources (AIGDER) represent a novel resource form that leverages generative foundation models to meet user needs. Single-agent approaches suffer from weak planning and narrow perspectives, while multi-agent systems face challenges in educational contexts including human-machine intention misalignment, coordination conflicts, and unstable output quality. To address this, this paper introduces Shared Mental Models (SMM) theory, mapping its two-dimensional structure onto multi-agent technical elements to construct a multi-agent-supported AIGDER development model. Built on a core workflow of demand analysis, resource generation, application evaluation, and resource optimization, the model adopts SMM’s three-stage evolution of initial formation, consensus adjustment, and fluency as its operational mechanism, driving cognitive alignment and coordination norm consistency to form a human-machine collaborative development loop. Implementation safeguard strategies are further proposed across four dimensions of technical architecture, semantic states, process control, and quality optimization to enhance stability, controllability, and sustainable evolution, offering reference for both theoretical research and practical application in AIGDER development.