Context-awareness can be used to simplify a user's understanding of, and interaction with, interactive systems. In effect, through adaptation, context-aware systems can migrate complexity away from the user and into the system (or agent). However, the incorporation of context-awareness raises a number of issues. For example, users are required to trust the behaviour of the system's intelligence and this requires the system to have predictable behaviour and the ability to successfully and consistently preempt the user's goal. Unfortunately, the agent may incorrectly preempt the user's goal, owing to either flawed intelligence or to incorrect or out-of-date contextual information. In such circumstances the user is likely to feel frustration because the system will either appear overly prescriptive or, worse still, present incorrect results. This paper considers these issues, a number of which are described in anecdotal form, based on our experiences in developing and evaluating the context-aware GUIDE system.