Adaptive efficiency is the product of an evolutionary process in which certain types of learning prevail, locking societies in vicious cycles of underdevelopment or unlocking the path to prosperity. Moving development studies from a theory of choice to a theory of change, this research posits that learning reinforces path dependent processes of underdevelopment and identifies barriers that block the reflective learning types that can unlock society from such patterns. With data about the ability to learn of urban development projects, the empirical analysis studies learning fitness with a model of Evolutionary Game Theory, combined with a process tracing method in case studies. The results present evidence that hierarchical power relations and complexity increase path dependence to unreflective learning and that vicious cycles of underdevelopment are reinforced by ambiguity, defensiveness and misperceptions of dynamic complexity. The conclusion recommends to policy makers a focus shift from just policy design to learning processes.