Most serverless platforms model applications as workflows that specify how each function is invoked (e.g., by events, predefined rules). However, these function orchestration services do not consider how data is exchanged between functions. When a function is not entirely and immediately consumed by a subsequent function, developers must manually manage function interactions and data exchanges using complex logic and other tools like a message broker or shared storage. To enable data-centric function orchestration, Yu et al. present Pheromone, a low-latency serverless platform that provides a rich set of data trigger APIs that developers use to specify when and how intermediate data are consumed by downstream functions.