There has been a dramatic rise in the use of the local delivery model for development interventions, where local agents are hired as intermediaries to target interventions to potential beneficiaries. We study this model in the context of a standard agricultural extension intervention in Uganda, in a setting where communities are highly politically polarized. We use a two-stage field experimental design. In the first stage, we randomize the delivery of the intervention across communities. In the second stage, in each community, we randomly choose one delivery agent out of two potential candidates. This design yields exogenous variation in social ties to the actual delivery agent as well as to her counterfactual. We reveal the key role the relationship between the actual and counterfactual delivery agents plays for how the intervention unfolds in these communities. The number of farmers targeted by the delivery agent, whether their own social ties are targeted, and whether they engage in pro-poor targeting all depend on whether actual and counterfactual delivery agents are aligned or divided in their political identities. As counterfactual agents play no formal role in the intervention, we interpret their in‡uence as a social incentive provided to the delivery agent, varying with the political alignment between the two. We document the impact social incentives have for resource allocation, inequality, and welfare, and narrow down the structure of social incentives consistent with all aspects of delivery agent behavior. Finally, we discuss the implications of our findings for the design of the local delivery model.