In this talk I will give an overview of the challenges of numerical weather prediction, illustrated by the pioneering work of Lewis Fry Richardson during the First World War. I will describe ongoing work in collaboration with the UK Met Office on a new dynamical core design motivated by the need to exploit greater parallelism in order to efficiently harness the power of future supercomputers. This work has led to the development of Gusto, the dynamical core toolkit built on top of the Firedrake finite element library, that provides a tool for rapid prototyping of new algorithms. I will finish by describing our plans for the future, in particular the need for increasing parallelism which is driving developments in time-parallel algorithms which pose challenges for hyperbolic problems such as those used in numerical weather prediction.