With the deep integration of generative artificial intelligence technology and digitalized classrooms, AI agents driven by multimodal large models have gradually become a key carrier for teaching activities. Referring to the component structure of AI agents and combining relevant research findings on AI agents, this paper designed an AI agent teacher, elaborated on its technical architecture, and explored its teaching implementation. On this basis, relying on the human-machine collaborative “dual-teacher co-teaching” model and guided by the dual-process cognitive development theory and constructivist teaching theory, this paper further constructed the human-machine collaborative dual-teacher model embedded with AI agent teacher, and introduced the relevant situation of its deployment and application. To evaluate the application effect of this human-machine collaborative dual-teacher model embedded with AI agent teacher, this paper conducted the teaching application of this model in “Computer Architecture”, a professional course for undergraduate students majoring in computer-related disciplines. Through a comparative experiment on teaching application effects, this paper found that this model enhanced students' academic performance and cognitive abilities, and also had a certain positive effect on their learning motivation. The research in this paper provided theoretical guidance for the development of AI agent teachers and offered an effective pathway for applying AI agent technology to teaching practice, and could help to promote the implementation and paradigm transformation of human-machine collaborative teaching.