A preprint on arXiv describes a large-scale field experiment designed to uncover how AI agents affect teamwork and performance. The researchers built a platform called Pairit specifically to study human-AI collaboration, allowing them to run a controlled experiment in a realistic setting.

The study randomly assigned 2,234 participants, providing a robust sample for measuring the impact of AI agents. Rather than simply measuring gains, the work focuses on the mechanisms that drive productivity and performance improvements.

The abstract is truncated, so specific findings are not yet visible in the summary. The paper's contribution lies in its experimental design and the development of a dedicated platform for this type of research.