Mixed reality (MR) and AI-powered pedagogical agents have been increasingly explored in STEM labs context as a means of providing hands-on practice with virtual teaching assistants. However, teaching microelectronics presents unique challenges because learners must integrate conceptual knowledge, procedural skills, and visual–spatial understanding to reason about multilayer chip structures and how fabrication processes optimize performance. We introduce ChipXR, a mixed-reality learning environment developed through a design-science approach and multistakeholder participatory design involving instructors, teaching assistants, students, education researchers, and HCI researchers to systematically refine interaction design and reduce potential cognitive load. The final prototype integrates interactive lecture modules for conceptual understanding, embodied MR activities for procedural skill development, and a LEGO-based building task to foster visual–spatial reasoning, all supported by a GPT-powered agent that explains the principles behind each activity. In a controlled study comparing MR+AI agent, MR-only, and traditional video-based lecture, we found no significant differences in conceptual quiz scores, but the MR+AI condition showed trends associated with visual–spatial understanding and perceived knowledge gains. A deployment in an introductory advanced semiconductor packaging course, supplemented by think-aloud sessions and focus groups, further contextualizes these findings. Together, these results suggest how MR+AI systems can be purposefully orchestrated to target complementary knowledge types and inform the design of future microelectronics curricula and other complex STEM domains requiring conceptual, procedural, and spatial reasoning.