I present a neural network model for the spontaneous emergence of enterprises following a dynamic approach to firm formation in the tradition started by Adam Smith and further pursued by Joseph Schumpeter. I suggest that the “natural propensity to truck and barter” is the observable behaviour of self-interested economic actors who are continually exposed to and confronted with an environment complexity, which transcends their limited cognitive and computational capabilities. In their learning process they build networks of relations with other agents. It is this interaction among heterogeneous agents that often leads to the formation of successful organizations that completely solve the original problem. Firms can be seen simultaneously as the result of the entrepreneurs induction process, but also as the essential instrument for the elaboration of a solution to the problem. Increasing returns to scale and market size find in this framework a very natural representation. The role of competition for the nurturing of efficiency, and the issue of protection of infant industries can be tackled by this model.