The paper studies a simple two-period principal/agent model in which the principal updates the incentive scheme after observing the agent's first-period performance. The agent has superior information about his ability. The principal offers a first period incentive scheme and observes some measure of the agent's first-period performance (cost or profit), which depends on the agent's ability and (unobservable) first-period effort. The relationship is entirely run by short-term contracts. In the second period the principal updates the incentive scheme and the agent is free to accept the new incentive scheme or to quit. The strategies are required to be perfect, and updating of the principal's beliefs about the agent's ability follows Bayes' rule. The central theme of the paper is that the ratchet effect leads to much pooling in the first period. First, for any first-period incentive scheme, there exists no separating equilibrium. Second, when the uncertainty about the agent's ability is small, the optimal scheme must involve a large amount of pooling. The paper also gives necessary and sufficient conditions for the existence of partition equilibria and looks at the effect of cost uncertainty.