The World Wide Web has evolved drastically over the past decade – and the proliferation of Web APIs has turned it into the middle- ware of choice for most distributed systems. The recent focus on hypermedia-driven APIs together with initiatives such as the Web of Things and Linked Data are now promoting and advancing the development of a new generation of dynamic , open , and long-lived systems on the Web. These systems require agent-based solutions to the point that Web researchers have started to build autonomous systems on their own. It is thus both timely and necessary to in- vestigate and align the latest developments in Web research and multi-agent systems (MAS) research. In this paper, we analyze in hindsight the factors that hindered the widespread acceptance of early Web-based MAS. We argue that the answer lies equally in a lack of practical use cases as well as the premature development and alignment of Web and agent technologies. We then present our vision for a new generation of autonomous systems on the Web, which we call hypermedia MAS , together with the research opportunities and challenges they bring.