Starting from the separation between formal and real authority, the paper considers a hierarchical relation where delegation of control can be used as an incentive mechanism. It shows that delegation is optimal when parties' interests are neither too divergent nor too close. In that case superiors do not need to monitor too closely, curbing the subor- dinates' initiative and subordinates have no incentives to free ride on the superiors' information. The analysis is then extended to a multi- period setting. If subordinates acquire expertise on the job (learning by doing), effort has the nature of a specific investment whose intrinsic value is higher the longer the subordinate's working life. Therefore, a policy of gradual delegation is possible, where agents accept weaker in- centives at the beginning of their career, requiring stronger incentives over time as the value of their knowledge diminishes.