Generative AI has made it faster to find promising molecules in early drug development, but turning these into real medicines remains a challenge. This chapter explores key issues, including the limits of ADMET prediction, generalization, and reproducibility. It presents case studies on kinase inhibitors, antimicrobial peptides, and COVID-19 repurposing to highlight where current methods succeed and where they fall short. The chapter also addresses ethical and regulatory topics like transparency and data privacy. Finally, it offers practical strategies—such as explainable AI, federated learning, multi-omics integration, closed-loop automation, and community standards—to help bridge the gap between computer predictions and real-world outcomes, supporting broader adoption of AI in drug development and regulation.