With the rapid penetration of generative artificial intelligence in educational scenarios, AI programming assistants have gradually been applied in STEM programming classrooms. However, whether they enhance or weaken the development of students' computational thinking remains an unclear conclusion. Therefore, in technology-enhanced STEM education, how to leverage the enabling value of AI while maintaining the core boundaries of computational thinking development has become a key issue that current student programming education urgently needs to address. This study, based on the theoretical framework of technology-enhanced education and STEM education, adopts a quasi-experimental research design. It selects undergraduate students from different universities as research subjects, sets up an AI-assisted learning group and a traditional programming teaching group, and through a one-month teaching intervention, systematically explores the dual impacts of AI programming assistants on the computational thinking of undergraduate students. This study uses methods such as computational thinking scale, programming work analysis, classroom observation, and interviews to conduct evaluations in four dimensions: abstract thinking, algorithm design, problem decomposition, and debugging and error correction. The expected results aim to reveal the dual value and boundary conditions of AI tools in university programming education and propose reasonable teaching strategies that balance efficiency and thinking cultivation, providing empirical evidence and practical references for the high-quality development of STEM education under technology empowerment.