In the new millennium, work involves an increasing amount of tasks that are knowledge-rich and collaborative. We are investigating how semantics can help on both fronts. Our focus is scientific work, in particular data analysis, where tremendous potential resides in combining the knowledge and resources of a highly fragmented science community. We capture task knowledge in semantic workflows, and use skeletal plan refinement algorithms to assist users when they specify high-level tasks. But the formulation of workflows is in itself a collaborative activity, a kind of meta-workflow composed of tasks such as finding the data needed or designing a new algorithm to handle the data available. We are investigating "organic data science", a new approach to collaboration that allows scientists to formulate and resolve scientific tasks through an open framework that facilitates ad-hoc participation. With a design based on social computing principles, our approach makes scientific processes transparent and incorporates semantic representations of tasks and their properties. The semantic challenges involved in this work are numerous and have great potential to transform the Web to help us do work in more productive and unanticipated ways.