There is now ample experimental evidence that speciose assemblages are more productive and provide a greater amount of ecosystem services than depauperate ones. However, these experiments often conclude that there is a higher probability of including complementary species combinations in assemblages with more species, and lack a priori prediction about which species combinations maximize function. Here I report the results of the first experiment that manipulates the evolutionary relatedness of constituent plant species across a richness gradient. I show that assemblages with distantly related species contributed most to the higher biomass production in multispecies assemblages, through species complementarity. Species produced more biomass than predicted from their monocultures when they were in plots with distantly related species, and produced the amount of biomass predicted from monoculture when sown with close relatives. This suggests that, in the absence of any other information, combining distantly related species in restored or managed landscapes can serve to maximize biomass production and carbon sequestration, thus merging calls to conserve evolutionary history and maximize ecosystem function.