Abstract: Microbial communities carry out essential functions for global climate, human health, and industry. Predicting their behaviors has proved challenging due to the tremendous amount of functionally relevant diversity observed at all levels of taxonomic resolution in nature. Yet, surprisingly, some ecosystems appear to be “coarse-grainable,” meaning some of their properties can be predicted with simple models ignoring much of the complexity. Understanding and harnessing ecosystem coarse-grainability would enable new approaches to prediction and design of these complex systems. However, we lack a theoretical understanding of this effect, or even the terminology to quantify or investigate it, limiting progress. I will present a mathematical framework for defining and investigating coarse-grainability based on the capacity of simple models to predict community-level functional properties. Applying this framework to data (1) reveals surprising evidence that increasing community richness can make simple models more predictive rather than less, and (2) underscores the need for a new kind of ecological theory, shifting the focus from simple and complex ecosystems to classifying simple and complex *properties* of ecosystems.