Abstract: Smart farms are revolutionizing traditional agriculture by integrating advanced technologies to enhance productivity and sustainability. Despite their potential, many existing systems are highly customized, difficult to replicate, and rely heavily on human intervention. Drawing inspiration from computer operating systems, we introduce the Farm Operating System (FOS). FOS adopts a crop-centric view point and leverages artificial intelligence (AI) for agricultural activities in the aspects of real-time situation awareness, data-driven planning, and automated decision-making. Under its guidance, we propose Beehive, an agent-based implementation of FOS. Beehive shifts the focus of farm from data visualization to crop growth, decomposing agricultural tasks into loosely coupled agents, each of which is realized by a modular, pluggable agent that can be installed as needed. A task is accomplished by orchestrating these agents, with the orchestration generated automatically by a large AI model. Beehive enhances adaptability while reducing system migration costs, addressing the challenges of smart farm adoption and positioning farms to evolve with future technological advancements.