We consider two settings on in which isolated agents perform costly experiments to infer the quality of a new technology and another in which the same agents are embedded in a network and can observe the outcomes of neighbors. We ask what is the best network structure for learning.