The literature is increasing on how to prioritize climate-smart options with stakeholders but relatively few examples exist on how to co-design climate-smart farming systems with them, in particular with smallholder farmers. This article presents a methodological framework to co-design climate-smart farming systems with local stakeholders (farmers, scientists, NGOs) so that large-scale change can be achieved. This framework is based on the lessons learned during a research project conducted in Honduras and Colombia from 2015 to 2017. Seven phases are suggested to engage a process of co-conception of climate-smart farming systems that might enable implementation at scale: (1) "exploration of the initial situation," which identifies local stakeholders potentially interested in being involved in the process, existing farming systems, and specific constraints to the implementation of climate-smart agriculture (CSA); (2) "co-definition of an innovation platform," which defines the structure and the rules of functioning for a platform favoring the involvement of local stakeholders in the process; (3) "shared diagnosis," which defines the main challenges to be solved by the innovation platform; (4) "identification and ex ante assessment of new farming systems," which assess the potential performances of solutions prioritized by the members of the innovation platform under CSA pillars; (5) "experimentation," which tests the prioritized solutions on-farm; (6) "assessment of the co-design process of climate-smart farming systems," which validates the ability of the process to reach its initial objectives, particularly in terms of new farming systems but also in terms of capacity building; and (7) "definition of strategies for scaling up/out," which addresses the scaling of the co-design process. For each phase, specific tools or methodologies are used: focus groups, social network analysis, theory of change, life-cycle assessment, and on-farm experiments. Each phase is illustrated with results obtained in Colombia or Honduras.