This paper shows how the experiment of the Diatelic Project, a Continuous Ambulatory Peritoneal Dialysis monitoring system, has taught us rules concerning the creation of intelligent agents based on a similar architecture. In particular, we explain how a fuzzy discretization of continuous sensors allows for a great model simplification while keeping some good precision in the diagnosis. The adaptation of this architecture to an anaesthesia-monitoring problem is developed, highlighting the qualities and the drawbacks of this kind of models.