Expert agents, such as lawyers, play a prominent role in conflict resolution, yet little is known about how they affect outcomes. We construct a model that permits us to estimate the influence of agents and test whether the parties in a dispute face prisoner's dilemma incentives. Using eighteen years of final-offer arbitration data from New Jersey, we find the parties do significantly better when they retain agents and that the parties learn about this benefit over time. However, we also find that the gain to using an agent is fully offset when the opposing party also hires an agent. Since agents are costly, this noncooperative equilibrium is Pareto inferior.