Web-based information work often involves juggling several independently developed pages, and the need to coordinate them frequently emerges only as a user browses. The arXiv abstract for Cling describes this as a problem with manual coordination: it requires repeated low-level actions, while explicitly specifying operations in advance is often impractical.
Cling is proposed as a middle path, enabling human-directed, in-situ, and small-step automation. Rather than asking users to define a complete workflow up front, the system supports incremental actions at the point of need, letting goals and coordination evolve alongside browsing.
Because the paper is a new preprint, its evaluation details are not available in the abstract. The main contribution appears to be a design philosophy that blends human control with automation for web tasks too dynamic for pre-specified scripts.