Abstract: With the rapid development of generative artificial intelligence technology, programming and engineering education are undergoing profound structural transformation. Large language models (LLMs), represented by DeepSeek, possess powerful capabilities in natural language understanding, algorithmic reasoning, and code generation, challenging the traditional teaching model centered on “programming implementation.” Taking the “FPGA image-processing remote experimental platform” as the research vehicle, this study explores systematic transformation pathways for experimental teaching empowered by LLMs. By embedding AI agents throughout the entire teaching process, a closed-loop teaching system of “AI-supported learning guidance – collaborative code generation – intelligent feedback analysis – innovative design demonstration” is constructed. The research achieves systematic reconstruction from three dimensions: teaching objectives, teaching activities, and teaching evaluation. Teaching practice using the case of “image denoising and edge-enhancement system design” shows that AI-empowered remote experimental teaching significantly improves students’ learning efficiency (shortening preparation time by an average of 35%), increases the code comprehension rate and the proportion of innovative designs (by 24% and 28%, respectively), and effectively reduces teachers’ repetitive guidance workload. The results indicate that LLM-empowered remote experimental teaching not only optimizes instructional processes, but also promotes a shift in teaching logic from “operation-driven” to “intelligence-driven,” providing a new paradigm for the digital and intelligent reform of engineering education.