The delivery of development programs is often delegated to local delivery agents (DAs). A common presumption is that, in the absence of financial rewards, social ties with potential beneficiaries might motivate the DA but also distort targeting. We provide evidence on the conditions underpinning the trade-off by means of a two-layered field experiment in the context of an extension program meant for the poorest farmers in Western Uganda. The first layer is a standard cross-village RCT that reveals significant gains in agricultural output and consumption in treated villages. The second layer randomizes the selection of the delivery agent between two elite farmers. This creates a counterfactual delivery agent (CA) for the agent and for her social ties. We find that social ties create bias but also improve coverage and that it is the rivalry between the agent and her counterfactual that creates this trade-off. The results have implications for the procedural rules used by governments or NGOs when recruiting DAs, and help explain why otherwise similar development often succeed in some communities and fail in others.