Is top-down organization design worth attempting at all, or should organizations simply let their members learn which patterns of interaction are valuable by themselves, through a bottom-up process? Our analysis of an agent-based computational model shows that weak enforcement of even a randomly selected formal structure in a top-down manner can usefully guide the bottom-up emergence of networks of intraorganizational interactions between agents. In the absence of formal structure, interactions are prone to decline within organizations, because maintaining interactions requires coordination but breaking them does not. Formal structure regenerates the network of interactions between agents, who can then learn which interactions to keep or discard. This “network regeneration effect” of formal structure offers a rationale for the importance of top-down organization design, even if the design is limited in accuracy and enforcement.