This paper will present the results of a case study on radio use conducted with people with dementia (PwD). We further highlight design implications regarding a long-term assistive environment for PwD. Introduction: Dementia is a progressive neurodegenerative syndrome that mainly affects higher cognitive function in people at their older ages. As a result, PwD often have difficulties carrying out everyday activities. But PwD also develop psychological deficits such as anxiety or inferiority complex. We have conceived a framework to offer adaptive long-term assistance to PwD by taking their current state, context and intentions into account. Part of this development is an approach we call benevolent manipulation: Regardless of whether the right actions are taken or not, an assistive device or service will (ideally) always meet a person’s actual needs. This requires sophisticated techniques for action recognition and tracing, intention detection, and aid effect optimisation. Music is very important to people with dementia. But instead of developing yet another new device, we pursue the strategy of making familiar devices more intelligent. Methodology: We conducted a case study with eight participants with moderate to severe dementia. In semi-structured interactive interviews, they were asked to turn a retro-style radio (1950s to 60s) on and off, change the volume and the station. A human wizard controlled the radio remotely in order to assure that all (suggested) intentions were met. Participants were video-taped and audio-recorded. Main results: Among the main findings are: 1) Participants liked the look and feel of the radio and were quickly engaged, 2) most of them felt self-confident and happy as there were no problems during use, 3) there was a considerable variability in the actions different participants took for the same task, 4) but only little differences were observed during use within participants after a one week break. Conclusion and future work: To conclude, we have shown that, by introducing benevolent manipulation to radio use, we can lower barriers of interaction, motivate PwD, and stimulate positive emotions. In the future, we plan to replace our wizard by an affordable system that can intelligently adapt to the particular needs of PwD. This will help to offer more (felt) independence, self-esteem, and so improve their (perceived) quality of life. In more detail, based on formal descriptions of the actions we have observed, and on non-intrusive / -disturbing sensor technology, we can infer what the behaviours and intentions are in a given context. To this end, the “raw sensor data” have to be linked to abstract models of behaviors in order to obtain symbolic events that can be evaluated using efficient plan recognition techniques. Context and intention information can then be used to select the most appropriate assistance while considering individual perceptual, cognitive, social and contextual constraints. These constraints will be fed into a cognitive simulation (e.g., in ACT-R) whose main purpose is to probe and optimise aid selection. As a result, large parts of the system can run in an autonomous closed loop.