We strongly believe that agent-oriented approaches to system development come with a natural level of abstraction and therefore have something valuable to offer. However, in doing so, any comprehensive agent-oriented methodology necessarily has to be grounded in issues and solutions of relevance in contemporary research and development areas such as Grid computing and autonomic computing - in order to realize the visions of ambient intelligence. Current efforts of AOSE, however, mostly focus on traditional methods of software development, provides implementations of stand-alone agent systems, or isolated experimental platforms. These efforts are, of course, worthwhile in themselves but have clear limitations when it comes to their contribution and fulfillment of visions such as ambient intelligence. Consequently, in this chapter we introduce the methodological approach of online engineering. As such, this methodology has explicitly been designed to meet what we conceive as the major challenges and limitations in contemporary approaches of AOSE. In fact, we argue that these limitations primarily are due to a strong focus on current practice in software engineering, rather than on engineering of grounded open computational systems. In this respect, online engineering provides us with the models, methods, and tools to facilitate the necessary transition from programming of abstract machines towards development of grounded physical systems, e.g., from software engineering to engineering of open computational systems.