Agents have foresight when they receive information about a random process above and beyond the information contained in its current and past history. In this paper, we propose an information-theoretic measure of the quantity of foresight in an information structure, and show how to separate informational assumptions about foresight from physical assumptions about the dynamics of the processes itself. We then develop a theory of endogenous foresight in which the type of foresight is chosen optimally by economic agents. In a prototypical dynamic model of consumption and saving, we derive a closed-form solution to the optimal foresight problem.