In this new era of AI with diverse hardware accelerators such as GPUs and quantum circuits, achieving system-wide robustness requires tackling issues throughout all system layers, spanning from software applications to hardware components. My research is to enhance the robustness of heterogeneity-enabled AI systems by reinventing software testing and analysis techniques via leveraging full-stack insights and advanced AI capabilities. I have completed one research project and have collaborated on a couple of others at the application and language levels. As the next steps, I will explore (1) holistic regression testing to prioritize test inputs associated with system-wide changes and (2) full-stack analysis to optimize computing resource allocation and reduce hardware reliance by analyzing application characteristics and using alternative resources in tandem.