Abstract: Recent advances in artificial intelligence (AI) tools, including large language models (LLMs) and physics-informed neural networks (PINNs), are reshaping the teaching of computational mechanics from a conventional mode based on classical numerical methods and hand-coded programming toward closer integration between classical methods and modern AI tools. To explore how AI tools can be incorporated into undergraduate mechanics education, this study uses the compulsory course Engineering Design and Optimization Methods as the teaching context and develops a three-level instructional framework. First, LLMs are employed throughout the course to assist with question answering, programming, and documentation retrieval, while their use is regulated through an “AI usage log” mechanism. Second, teaching examples motivated by multi-objective optimization of ball grid array (BGA) solder joints and parameter identification in phase-field fracture are used to demonstrate neural-network surrogate modeling, Bayesian optimization, and error verification with parameterized teaching response models. Third, emerging AI methods, including PINNs and Kolmogorov–Arnold networks (KANs), are introduced to broaden students’ understanding of AI for Science and interpretable modeling. These teaching practices help clarify the roles and application boundaries of AI tools and establish a progressive instructional framework. The study may provide a reference for engineering talent development, smart-course construction, and curriculum reform in computational mechanics in the AI era.