Organisms not only respond to their environment but also influence the availability of resources and change environmental conditions. Hence, the impacts of organisms on their environment shape the selective regimes that drive, on ecological time scales, the assembly of ecological communities and, on evolutionary time scales, diversification. Recent studies have drawn attention to the fact that feedbacks between organisms and the environment can prevent or induce catastrophic transitions in ecosystem states and argue that climate change increases the likelihood of such catastrophic regime shifts. Ecologists have very limited ability to predict the likelihood of such regime shifts or the properties of the ecosystems that assemble after such collapses. This is because ecology does not have a theory of ecosystem assembly, nor does it have an established way of translating such a theory into models capable of predicting future ecosystem states. Without knowing these potential endpoints, we cannot develop strategies for coercing ecosystems into desired states, severely constraining our capacity to mitigate climate change and climate change impacts. This paper outlines a roadmap for developing a theory of terrestrial ecosystem assembly. Recent progress in dynamic global vegetation modelling and community assembly provides a useful foundation for a theory of ecosystem assembly. Environmental filtering and limiting similarity are key principles, but to be useful, they need to be linked to resource consumption and environmental modulation, and be more strongly constrained by biophysics and the trade-offs defined by biophysical principles. Such a theory recognises that ecological and evolutionary history ensures that many different ecosystem assemblies are possible at any given point in space and time.