This presentation focuses on the prototyping of a new kind of decision support tool (DST) by a French “technical institute”, a para-public structure where engineers and technicians cooperate for applied agronomic research and dissemination within agricultural worlds. This kind of DST relies on a specific mathematic tool, the “directed acyclic graph”, in which each variable for modeling agricultural practices (i.e. “potassium intake”) is linked to parent (causes) and child (consequences) variables through probability tables. Within the Institute, his kind of software is deemed to be a revolution from the local paradigm of randomized experimentation; it aims to model agroecological practices and their consequences on agricultural production. We conducted ethnographic observation of conception sessions involving agronomic experts (on water, soil, plant, disease, etc.), during which all of them had to discuss the variables that may have an effect on the phenomenon being modelled (crop yield, water supply, etc.), and quantify their relations. How does this heterogenous prototyping configuration shape knowledge production, and attempt to embed agronomic knowledge into digital infrastructure? Our proposal therefore revisits the observations of the first laboratory ethnographies (Latour, Woolgar and Salk, 1979; Knorr-Cetina, 1999), by raising the question of digital technology’s role in the mobilization of experimental practices and arenas of knowledge production. We document prototyping activities not only as forms of knowledge production or design management, but also as a method of interweaving innovation trajectories (agroecology and digital in this case) and forms of organisation or collective action.