Many development programs that attempt to disseminate improved technologies are limited in duration, either because of external funding constraints or an assumption of impact sustainability; but there is limited evidence on whether and when terminating such programs is efficient. We provide novel experimental evidence on the impacts of a randomized phase-out of an extension and subsidy program that promotes improved inputs and cultivation practices among smallholder women farmers in Uganda. We find that phase-out does not diminish the use of either practices or inputs, as farmers shift purchases from NGO-sponsored village-based supply networks to market sources. These results indicate short-term interventions can suffice to trigger persistent effects, consistent with models of technology adoption that emphasize learning from experience.