Clinical trials are the ultimate and most expensive tools in the development of medical products. Their designs have changed little since the advent of randomization in the 1940s. Confirmatory trials are large and slow and take years to answer a single question. They are completely out of sync with modern biology that is changing our understanding of disease and its heterogeneity not yearly but weekly. Building a design is making a decision. Modern innovations are going back to the drawing board, borrowing from the theory of bandit problems and asking how we can design clinical trials with the explicit goal of delivering better therapy to patients. These new designs incorporate randomization but in a much more efficient and dynamic fashion. They address many questions of clinical and biological importance, including those related to disease heterogeneity. They lead to different approaches for handling and modeling control therapy. They focus on the future, predicting each patient's disease course and predicting each therapy's success in the present trial and in future trials. Evaluating the performance of such complicated designs requires simulation. I'll describe actual clinical trials that are incorporating decision-analytic concepts.