Research questioning ability is an important foundation for the cultivation of postgraduates’ academic capabilities, and has a significant influence on the establishment of research directions and the improvement of research quality. However, current postgraduates generally face problems such as unclear concepts of research questions, lack of questioning strategies, and low quality of questions. Therefore, this paper first designed research questioning activities supported by multi-agents and developed the multi-agents research questioning tool “ZhiXue Yanyou”. Subsequently, this paper adopted a combination of questionnaire surveys, expert scoring, and semi-structured interviews to conduct the pre-test and post-test quasi-experiment with unequal groups. The results showed students had a relatively high overall satisfaction with the use of “ZhiXue Yanyou”; multi-agents can effectively improve students’ research questioning ability, assisting them in mastering research question theoretical knowledge, applying questioning strategies, and improving question quality, but their role in supporting high-level cognitive processes such as question evaluation, judgment of question importance, and in-depth analysis of professional fields was limited. Finally, this paper conducted in-depth discussions from three aspects: adaptability of multi-agents, cognitive gap, and development of literacy, highlighting the application potential of multi-agents in the cultivation of postgraduates’ research abilities, and providing valuable references for the optimization design and practical application of AI educational tools in the future.

Large language model agents for education is a new generation of generative artificial intelligence technology, which could utilize the large language model to decompose and plan tasks, invoke tools and knowledge bases, and complete complex and diverse educational tasks. However, there are also problems such as shallow understanding of its technical principles and new features, incomplete exploration of application status, and unclear educational effectiveness. Therefore, this paper discussed the development situation of large language model agents for education from a technical perspective, including the design of its technical architecture and the analysis of the advantages and disadvantages of its mainstream development framework and platform. Then, in order to present the application status of large language model agents for education, this paper introduced 20 typical large language model agents for education at home and abroad for comparative analysis, extracted new features of large language model agents for education, summarized the roles of large language model agents for education in four application scenarios of teaching, learning, management, and evaluation, and introduced the application cases of large language model agents for education in different scenarios. Finally, combined with the opportunities and challenges of large language model agents for education, this paper looked forward to its future prospects. The research of this paper could promote the development and innovative application of large language model agents for education, which was conducive to promoting the digital transformation of education and the high-quality development of education.