Despite major advances in high-throughput genomics and cell biology, a predictive understanding of human cell diversity and tissue function remains out of reach. A central challenge lies in integrating heterogeneous measurements across molecular, cellular, and tissue scales, from genome regulation and chromatin architecture to spatial organization, cell-cell interactions, and higher-order tissue structure. Recent advances in AI and machine learning, particularly generative and agentic AI, offer new ways to model cellular systems across these scales, learn biologically grounded representations from increasingly rich multimodal data, and generate predictions and hypotheses about complex cellular processes. I will present our recent work on modeling gene regulation and genome organization from single-cell epigenomic data, as well as cellular interactions and tissue architecture from spatial molecular profiles. I will also discuss how generative models and AI agents can extend these approaches toward predictive modeling and iterative scientific discovery. Together, these approaches provide new insights into how cellular states are established, coordinated, and perturbed in health and disease.