There is a new lab tool for studying how people and AI agents actually work together, not just chat.
Researchers have released Pairit, an online platform for designing and running live experiments on human-AI collaboration. A single YAML file lets a researcher declare the whole experiment: pages, routing, randomization, matchmaking, chat, shared workspaces, server-hosted AI agents, surveys, timers, and custom HTML components. Any mix of humans and AI agents can then join a live session where they communicate and act on shared tasks together. The platform has already been used in multiple live deployments, including peer-reviewed published studies, capturing detailed traces of communication, negotiation, and joint work between human-AI pairs.
The gap Pairit targets is real. Existing tools handle human-to-human sessions or real-time human-AI chat, but few let researchers set up protocols where AI participants both talk and act on shared work in one auditable setup. As companies push AI agents into actual workflows, not just Q&A boxes, understanding how mixed human-AI teams coordinate and delegate becomes a real organizational design question, not a philosophical one.
What is genuinely useful here is the auditability. A YAML config file is a reproducible, shareable record of an experiment's structure, which matters for a field that has struggled to standardize how it studies human-AI interaction. Whether Pairit becomes a common research substrate or a niche tool used by its own authors is the open question, and one paper's worth of deployments is not enough to answer it yet.