In this article we discuss the dynamics of organizational change when agents have heterogeneous initial conjectures and do learn. In this framework, conjectural equilibrium is defined as a steady state of the learning process, and all the adjustment occurs in disequilibrium. We discuss the properties of the system under different “rationality” assumptions, and using well-known learning algorithms. We prove analytically that multiplicity of equilibria, and failure of good organizational routines, cannot be ruled out: better, they are fairly probable. Stability is a crucial matter: it is shown to depend on initial conjectures. Finally, learning does not necessarily select the best.