We study the evolutionary properties of decision processes. We show that in the presence of harmful mutations, a population of agents who possess decision architectures with hierarchically organized modules will have a strictly higher asymptotic growth factor than a population of agents with decision architectures in which the modules are fully connected. We also show that under imperfect precision of the architectures and cyclical fluctuations in the environment, internal conflict within agents and behavioral heterogeneity across agents are properties of evolutionary equilibrium. Finally, we show that economic models of multiple decision processes can be represented as examples of modular hierarchies.