We provide a theory of the role of discretion in agency. We do so by extending the standard principal-agent framework to include complex information and trust. With these two new ingredients we find that: optimal agency contracts may be incomplete; they do not necessarily use all available, payoff-relevant information (the sufficient statistic result fails); and inefficiencies may persist even though inference about the agent's action choice is perfect. We show that an optimal response to complex information in agency relationships is for the contracting parties to choose exceedingly simple contracts---a fixed wage plus a discretionary bonus, for instance. In our model, a main function of agency contracts (in addition to the provision of incentives) is to protect the agent from possible opportunistic behavior of the Principal.