Distributed systems are at a watershed due to their increasing complexity. The heart of the problem is the extreme level of heterogeneity exhibited by contemporary distributed systems coupled with the need to be dynamic and responsive to change. In effect, we have moved from distributed systems to systems of systems. Following on from this, middleware is also at a watershed. The traditional view of middleware is no longer valid (i.e. as a layer of abstraction, masking the complexity of the underlying distributed system and providing a high-level programming model). In practice, middleware is often by-passed with complex systems constructed in a rather ad hoc manner as a mash-up of a variety of technologies. The end result is that middleware is no longer sure of its form or purpose and this lack of a viable approach is a huge barrier to the emergence of areas such as smart cities and emergency response systems. This paper argues that there is a need to fundamentally rethink the middleware landscape related to complex distributed systems. The core contribution of the paper is a set of fresh perspectives, which lead us in turn to novel principles and patterns for middleware and subsequently to new styles of platform. These perspectives include a move to emergent middleware, seeking flexible meta-structures for distributed systems, and a step away from generic to domain-specific technologies. A number of case studies are also presented to demonstrate what this might mean for future distributed systems.