Against the general feeling that ‘CFD is solved’, I offer a few remarks from the trenches. When performing large-scale simulations of geometrically and physically complex flows, the notion of reliable, cost-effective, widely applicable and easy-to-use CFD still appears more as fiction than reality. The present monograph addresses some major areas where progress is urgently required. These areas are listed in the order a typical CFD run proceeds: model import/creation, grid generation, flow solvers, mesh adaptation and quality control. Finally, with a view towards the coming decade, code, data and project management are discussed.