A fundamental pillar of behavior is seeking rewards. This involves both learning which actions lead to reward, and generating those actions. How is the striatum organized to support these two processes? The traditional model is that during learning, cortical inputs to the striatum are strengthened by dopamine-based reinforcement, and that that change in strength determines which actions are generated. To test this model, we have been systematically examining the contribution of dopaminergic and cortical inputs to the striatum of mice in order to understand how specific regions of the striatum perform these two computations. Our results suggest that dopaminergic inputs to different regions of the striatum contribute differentially to learning versus generating actions. In addition, we find that activity in cortical neurons that project to the ventral striatum provides information about prior actions that contributes to learning, but not to which action is generated. Some of these results, which we originally identified in a reinforcement learning task, seem to generalize to a social behavior assay. Together, our data suggest that inputs to specific striatal subregions are specialized to support each aspect of reward seeking: learning versus generating actions.