Cultivating students’ moral character, higher-order thinking skills, and human-AI collaborative literacy through personalized approaches represents a new direction for talent development in higher education. Adaptive generative learning (AGL), emphasizing learner agency, personalized processes, and generative outcomes, is a key talent development pathway for such development. This study adopted a mixed-methods approach integrating theoretical deduction, action research, and epistemic network analysis. This study (1) clarified the mechanism and processes of AGL among college students; (2) designed a multi-agent system (MAS) for AGL based on the “Cooperation-Hierarchical-Role” (C-H-R) collaboration mechanism; (3) constructed a MAS-based AGL model for college students to explore the promotional role of MAS in the AGL process; (4) developed a MAS using the COZE platform and conducted an empirical study with 71 sophomore students majoring in educational technology at S University. The results indicated that the MAS based on the C-H-R collaboration mechanism can significantly enhances college students’ AGL outcomes across cognition, ability, and experience.