The recent advances in technologies such as Artificial Intelligence, Internet of Things (IoT), automated drones, and embedded systems have involved tremendous changes in many industrial systems architectures. The shift from non-interoperable simple systems to complex systems that involve hundreds of devices, gathered by various networks happened at a fast pace, leading to situations where the operators themselves do not have a clear idea of what their system is, how to secure them and how to recognize faulty behavior, understand the causes of such a behavior and how to correct it. On the other hand, huge progress has occurred since the rise of Autonomic Computing twenty years ago. Indeed the initial promises of such systems - complete autonomy, self-healing properties to name a few - have been shown almost impossible to reach on real-life large-scale systems. However, a significant load of tasks may be offloaded to the operator by using some of those principles. Moreover, AI and machine learning recent advances greatly enhanced the capacity of a system to perceive and forecast its state. Intrusion Detection Systems (IDS) is one of the services that is at the junction of all those advances. AI may help malicious traffic detection on highly heterogeneous systems involving IoT, embedded systems, and more classic LAN. Human operators have to drive automatic actions against malicious patterns, while letting suspicious but honest traffic continue. Designing and implementing an efficient system in this case is an active research topic. In this paper, we consider the specific case of IDS for networks dedicated to agriculture, from the operational point of view: while much work has focused on new techniques to efficiently implement IDS on that setup, this paper aim to identify constraints and possible solutions to perform an operational deployment of such IDS.