Reading, as a fundamental pathway for knowledge acquisition, competence cultivation, and character shaping, plays a pivotal role in fostering innovative talents and enhancing national cultural soft power. However, current Primary School Chinese reading instruction suffers from fragmentation, superficiality, low cognitive engagement, and monolithic evaluation, failing to meet contemporary educational needs. To address these issues, under the guidance of reading instruction theory, project-based learning theory, and deep learning theory, this study adopted three research methods: literature research, theoretical deduction, and quasi-experimental research. Specifically, Firstly, it integrated and reconstructed the theories to form a theoretical framework for project-based deep reading in Primary School Chinese, systematically explaining the connotative characteristics and operational mechanism of project-based deep reading, and clarifying that project-based deep reading should take deep reading ability as the core goal and project-based learning as the main implementation method. Secondly, drawing on cognitive development theory, it conducted an in-depth analysis of the general process of project-based deep reading instruction, and combined the technical enabling role of AI Agent to construct the Agent-OPDE (Objective Guidance, Project-Driven, Deep Exploration, and Evaluation Feedback) instructional model for project-based deep reading in Primary School Chinese. Finally, it carried out teaching practice at Guangzhou N School, and verified the implementation effect of the model through multi-dimensional evaluation methods. Results demonstrate that this model effectively promotes students’ deep reading abilities, providing both theoretical foundations and a practical paradigm for Primary School Chinese reading instruction reform.