Information appliances, user interfaces, and context-aware devices are necessarily based on approximations of potential users and usage situations. However, it is not an unusual experience for developers that in some areas, appropriate approximations are extremely difficult to realize. Often these difficulties are not apparent from the beginning. Nevertheless, difficulties are rarely addressed in the pervasive computing literature as they appear to be peripheral compared to the technical challenges. In this talk, we argue that the field would largely benefit from addressing these issues explicitly. First, focussed discussions would help identify areas that have already shown to be difficult or even intractable in related disciplines, such as AI or CSCW. Second, it would help developers become aware of the difficulties and would allow them to deliberately circumvent such areas.