New personalized learning is oriented towards “capitalizing on strengths”, emphasizing the cultivation of students' unique and holistic personality traits. However, in large-scale teaching practices, it still faces potential risks such as narrow diagnostic dimension, imbalanced technological intervention, and superficial application. To promote the effective implementation of new personalized learning in large-scale teaching, this paper firstly took system theory and human-machine collaborative education theory as the theoretical basis and introduced AI agents to design a new personalized learning system based on AI agents. Subsequently, referencing classic personalized learning models, this paper constructed the new personalized learning meta-model based on AI agents, which included three elements of meta-learning, meta-path, and meta-diagnosis. Metalearning served as the foundation for learners to build an adaptive new personalized learning model, meta-path represented the basic units composing diverse learning paths for learners, and meta-diagnosis was the process of holistically evaluating learners' achievements in new personalized learning. Finally, this paper conducted an application experiment of the new personalized learning meta-model based on AI agents in the “Educational Big Data Thinking and Analysis Technology” course, and found that students adopting the meta-model performed better in data literacy and learning paths diversity. The research in this paper responded to the question of how to carry out new personalized learning based on AI agents from the perspective of humanmachine collaboration, providing new ideas for the implementation of large-scale individualized instruction, and also had an important enlightenment for promoting the transformation of the learning paradigm.

At present, the digital intelligentization of external society is continuously influencing the human subjects including teachers in the “three-dimensional space” through the way of knowledge combination. The complexity and diversification of knowledge acquisition channels make the cultivation of teachers’ teaching competency gradually point to the gerneral trend of “individuation”. One of the main directions of the classroom intelligent analysis large model training and the teachers’ competencies analysis and teachers’ professional development supported by artificial intelligence is to conduct vertical big model learning based on the case incremental data of the classroom intelligent analysis scenarios. Under this context, the paper firstly constructed the teachers’ teaching competency structure model based on the process perspective of ability formation and Spearman's two-factor theory. Subsequently, the intelligent analysis discrimination of classroom teaching behaviors and teaching competency was conducted, which constructed an analysis framework and training mechanism of teachers’ teaching competency based on the classroom intelligent analysis large model. Finally, taking a fifth-grade Chinese language teacher in an elementary school as an example, the mapping relationship between the teacher’s classroom teaching behaviors and teaching competency, as well as the framework of teacher’s teaching competency were preliminarily constructed. Through this research, the paper was aimed to provide strong support for the personalized training and intelligent assessment of teachers’ teaching competency.