Microbial communities in their natural environments are remarkably dynamic and heterogeneous. However, we still lack a framework to predict microbial behavior in such complex environments. Here, we use single-cell and microcolony microfluidic experiments to develop a mathematical model of the dynamics of bacterial cell responses, which captures the phenotypic diversity generated during antibiotic exposures. Then, we apply this model to analyze the short-term evolution of drug responses in continuous-culture evolution experiments. We show that loss of regulation of resistance, which is frequently observed in the clinic, can be favored during periodic exposure to large doses of antibiotics.