Intensive agriculture has led to several drawbacks such as biodiversity loss, climate change, erosion, and pollution of air and water. A potential solution is to implement management practices that increase the level of provision of ecosystem services such as soil fertility and biological regulation. There is a lot of literature on the principles of agroecology. However, there is a gap of knowledge between agroecological principles and practical applications. Therefore, we review here agroecological and management sciences to identify two facts that explain the lack of practical applications: (1) the occurrence of high uncertainties about relations between agricultural practices, ecological processes, and ecosystem services, and (2) the site-specific character of agroecological practices required to deliver expected ecosystem services. We also show that an adaptive-management approach, focusing on planning and monitoring, can serve as a framework for developing and implementing learning tools tailored for biodiversity-based agriculture. Among the current learning tools developed by researchers, we identify two main types of emergent support tools likely to help design diversified farming systems and landscapes: (1) knowledge bases containing scientific supports and experiential knowledge and (2) model-based games. These tools have to be coupled with well-tailored field or management indicators that allow monitoring effects of practices on biodiversity and ecosystem services. Finally, we propose a research agenda that requires bringing together contributions from agricultural, ecological, management, and knowledge management sciences, and asserts that researchers have to take the position of "integration and implementation sciences."