We develop a structural theory of beliefs and behavior, that relaxes the assumption of time-consistency in beliefs. Our theory is based on the trade-off between optimism, which raises anticipatory utility, and objectivity, which promotes efficient actions. We present it in the context of allocating work on a project over time, develop testable implications to contrast it with models assuming time-inconsistent preferences, and compare its predictions to existing evidence on behavior and beliefs. Our predictions are: (i) optimal beliefs are optimistic and time-inconsistent; (ii) people optimally exhibit the planning fallacy; (iii) incentives for rapid task completion make beliefs more optimistic and worsen work smoothing, while incentives for accurate duration prediction make beliefs less optimistic and improve work smoothing; (iv) without a commitment device, beliefs become less optimistic over time; (v) in the presence of a commitment device, beliefs may become more optimistic over time, and people optimally exhibit preference for commitment.