Agroecosystems are crucial to global food supply but are also a source of greenhouse gases. The vulnerability of agricultural systems to changes in climate mean and variability remains highly uncertain but has consequences both for food production and feedback to the climate. To examine these impacts, we partitioned changes in mean climate state and climate variability using an advanced weather generator. We used the parsed climate forcing individually and in combination to simulate U.S. Midwest agricultural regions using the Energy Exascale Earth System Model. We found that among the individual climate forcings, an increase in temperature mean has the largest impact and lowers stored carbon, plant productivity, and crop yield, and can cause local to regional cooling. We also find that the combined effect of mean and variability changes on carbon fluxes and pools was nonlinear. Our findings have implications for future food production and policies aimed at mitigating climate feedback.