This paper studies relational incentive contracts with persistent states in the presence of both moral hazard and information asymmetry. The optimal contracts are dynamic in which the agents are rewarded following a high output by moving to a higher continuation payoff in the next period. The stationary contracts as in Levin (2003) are no longer optimal. In a simple model, we show that when the principal is uncertain over the types of the agents, upon receiving a sequence of favorable outputs, the principal updates his belief about suppliers and give more favorable contracts to the agents. By using dynamic incentives the principal is able to eventually learn about the types of agents with time and can reach the first best. However, when cost parameters are different, the equilibrium dynamics may admit inefficient equilibria.