Researchers have released Kolab, a benchmark designed to study how humans and AI agents collaborate on long-horizon workflows. The work addresses a basic question: when people work with agents over extended tasks, how do the agents' contributions matter? The abstract indicates that the benchmark is built around measuring the value of iterative interaction rather than just final outputs.
According to the paper, for solving long-horizon tasks, humans place greater importance on ongoing interaction with AI agents, such as revising plans. This suggests that effective collaboration may depend less on a single strong answer and more on the agent's ability to engage throughout the workflow. Kolab appears intended to capture that dynamic in a reproducible way.
Because only the abstract is available, the specific tasks, metrics, and results are not detailed here. The benchmark's contribution, as described, is to shift evaluation toward the collaborative process itself.