There is a growing pressure to merge the study of organizations, on one hand, and cognitive science and AI, on the other. The underlying reason is twofold: first, classical theories of organizations are grounded upon models of rational action, largely consisting of means end reasoning and the principle of expected utility applied to planning. Secondly, and as a consequence, the game-theoretic background of rationality theories provided organizational theory with a highly formal methodology and a strongly deductive power. However, classical theories of organizations demand a deep revision, mainly because the very notion of rationality has been replaced by that of intelligent action. The revision process, however, should be bidirectional. On one hand, as is currently universally acknowledged, rationality is limited by cognitive constraints. Therefore, current AI studies have much to say to organizational theory. On the other hand, rational action is limited by social organizational constraints: organizations do not only emerge from interaction, they also pre-exist and constrain interaction. The very autonomy of real agents is limited by the structure of objective relations holding among them in a common world. Therefore, social and organizational theories have much to say to the study of both natural and artificial intelligence.