Renewable energy generation is inherently stochastic, and it rarely aligns with the periods of peak demand. Consequently, and despite continuous developments in storage technology, there is still a significant potential in using demand-side flexibility to balance generation and demand. The flexibility arises in many contexts, e.g., in the heating of buildings, and by shifting the demand, we can substantially reduce the need for infrastructure investments. However, it is not trivial to utilize the flexibility in a scalable manner. Energy customers are highly diverse (residential, commercial, industrial, etc.), and in most cases, their energy demand cannot be controlled directly. Furthermore, flexibility is both a dynamic and stochastic quantity. When the flexibility is used, it cannot be used at a later point in time as well. In this paper, we describe the Smart Energy Operating System, which is a framework for scalable exploitation of demand-side flexibility. It combines hierarchical forecasting with hierarchies of controllers and models. A key part of this framework is the Flexibility Function. It describes the energy demand of a flexible asset in response to a price signal, and it is continuously updated based on the actual demand. An aggregator can use it to predict the energy demand of the underlying flexible assets and participate in flexibility markets on their behalf. In other words, the Flexibility Function serves as a minimum interoperability mechanism (MIM). An important prerequisite is that markets must account for the dynamic and stochastic nature of flexibility, and we discuss current limitations and opportunities.