Appearing in the 1990s in the United States to analyze the rise and fall of Amerindian societies, agent-based modeling applied to archaeology has been developing for about fifteen years in Europe. In the field of Distributed Artificial Intelligence, ABMs offer the possibility of simulating the processes at work in the transformations that affect ancient societies by mobilizing multiple entities that interact with each other and with their environment, thus generating a dynamic. Simulations make it possible to test hypotheses concerning these processes to understand the trajectory of ancient societies, whose changes perceived by archaeologists and historians are interpreted by mobilizing different concepts (crisis, mutation, reorganization, etc.). Regardless of the results produced by these models, ABMs also have great heuristic virtues. The development of conceptual models and the computer implementation of these socio-environmental systems require a rethinking on the complex interactions between social processes and natural phenomena, on the different types of data to be integrated, on their articulation and the relevance of the parameters selected: this requires a close dialogue between human and social sciences, environmental sciences and computer sciences. The use of ABMs therefore appears to be a new challenge for the development of multidisciplinary studies on society-environment interactions. In this communication we will present in more detail the ROMCLIM and ROMPIRE-LPJmL models developed in RDMed and the ANR MICA projects to analyze the evolution of settlement, agricultural production and agrarian landscapes at different scales (southern Gaul, Roman Empire) between the Iron Age and the end of Roman antiquity, in relation to climate change. Based on the lessons learned from the development of these models and the analysis of their results, we will identify new avenues of research and perspectives for the development of ABMs applied to settlement, landscape archaeology and climate change.