Abstract-Nowadays, most effort in the area of context-aware systems goes into applications that process sensor data to proactively drive actuators. We share the concerns raised about such fully automated operation. Most notably, due to imperfect context inferences, actuating decisions are often contrary to the user's actual desires. Thus we focus on what we refer to as soft actuation: issuing low-key, non-verbal hints to the user, prompting him to optionally perform specific actuating actions. An actuating action consists in reaching to a nearby object and performing a simple manual operation on it. In this paper we describe the concept of soft actuation, position it with respect to related work, and identify relevant research challenges.