What should Earth system modeling, and computational science more generally, look like in the era of deep learning and generative AI? In this talk, I’ll describe our lessons from building NeuralGCM, our physics- and AI-based atmospheric model written in Python and JAX. I’ll explain the fundamental advantages of AI-based approaches, where they fall short, and how they can be effectively composed with physics-based models. I’ll also show how Google’s JAX framework is an incredibly powerful platform for building computational models.