Many development programs are based on short-term interventions, either because of external funding constraints or because it is assumed that impacts persist post program termination ("sustainability"). Using a novel randomized phase-out research method, we provide experimental tests of the effects of program phase-out in the context of a large-scale agricultural input subsidy and extension program operated by the NGO BRAC to increase the use of improved seed varieties and basic farming practices among women smallholders in Uganda. We find that while supply of improved seeds through local, BRAC trained women declined, demand does not diminish, and farmers shift purchases from BRAC to market sources, indicating a persistent learning effect. We also find no evidence of declines in the practice of improved and less costly cultivation techniques taught by the program. These results have implications for both efficient program design and for models of technology adoption.