Article dans une revue: Contemporary economic analysis is largely subject to rather bizarre schizofrenic syndromes. On the one hand, over the last thirty years or so, macro theories have tried to squeeze the interpretation of whatever aggregate dynamics down to some sort of decision-theoretic framework in which the increasingly mythical ''representative agent'' was doing all the action. Whatever statistical properties of the time-series, being it productivity and GDP growth, fluctuations, employment, investment, had to be explained as the equilibrium outcome of some sophisticated inter-temporal maximization exercise by such an agent. Dynamic Stochastic General Equilibrium models are the dominant genre in this spirit. On the micro side largely the opposite has happened. Empirical analyses drawing upon an increasing ensemble of micro longitudinal datasets have powerfully highlighted the ubiquitous, large and persistent heterogeneity in all dimensions of business firms' characteristics and dynamics one cared to look at.