Even though data is continually being produced by a wide range of services, harnessing this data in an effective manner at different levels of abstraction is challenging. In this vision paper, we consider how data streams representing real-world observations produced by geographically-distributed Internet of Things (IoT) devices can be used to adaptively fulfill a user's information needs. To solve this information need in a generic fashion, it is up to the underlying system to configure itself, split up processing across available resources, discover data sources, integrate available data, deal with uncertainty, integrate domain knowledge, etc. This is an interdisciplinary problem, and while many fields have considered the subproblems involved, this challenge requires a novel and integrated approach when these subproblems are put together. We propose Resource-Intelligent Stream Reasoning as an overarching solution to tackle this interdisciplinary problem. We define a novel set of challenges that capture the need for an integrated approach, show the limitations in the the current state-of-the-art and define the future directions in order to realize Resource-Intelligent Stream Reasoning.