Multicellular organisms depend on developmental programs to coordinate growth and differentiation from single cells, but the origins of development are unclear. A possible starting point is stochastic phenotypic variation generated by molecular noise. Given appropriate environmental conditions, noise-driven differentiation could conceivably evolve so as to come under regulatory control; however, abiotic conditions are likely to be restrictive. Drawing from an experimental system, we present a model in which environmental fluctuations are coupled to population growth. We show that this coupling generates stable selection for a single optimal strategy that is largely insensitive to environmental conditions, including the number of competitors, carrying capacity of the environment, difference in growth rates among phenotypic variants, and population density. We argue that this optimal strategy establishes stabilizing conditions likely to improve the quality and reliability of information experienced by evolving organisms, thus increasing opportunity for the evolutionary emergence of developmental programs.