Feature Request: Multi-Agent Shared Chats
I’d love to see ChatGPT support shared conversations with multiple AI agents or model instances working together in the same thread.
I discovered the usefulness of this accidentally while doing a long-term market-research project. I currently use one ChatGPT instance for exhaustive technical work—running large searches, testing strategies, validating results, and managing files—and another conversation for interpretation, critique, and designing the next experiment.
The combination is dramatically better than either alone.
A useful shared-chat version could allow the user to assign different roles, for example:
• Research/Execution agent — runs analyses, searches, calculations, files, and experiments.
• Critic/Auditor agent — checks assumptions, catches contradictions, identifies leakage or methodological problems.
• Synthesis/Strategy agent — interprets results, looks for deeper patterns, and suggests new hypotheses.
• User — directs the group and makes final decisions.
The agents should all have access to the same conversation context and authorized project files, but retain distinct roles and perspectives. They should be able to respond to and challenge one another directly rather than requiring the user to manually copy messages between separate chats.
The key benefit isn’t simply having “more AI.” It creates a built-in research team and adversarial review process. One agent can discover something, another can question whether the evidence really supports it, and another can recognize implications neither was explicitly searching for.
My current workflow already approximates this manually, and it has repeatedly caught errors and generated better research directions. The biggest weakness is the constant handoff and context duplication between chats.
Even a simple first version—two named AI participants plus the user in one project conversation—would be extremely powerful for research, engineering, coding, financial analysis, writing, and other complex work.
The feature I’m really asking for is not “multiple personalities.” It’s collaborative multi-agent reasoning inside one shared workspace.
I think people would find uses for it that are difficult to anticipate until they actually experience it.
That last line is the part I especially like, because that’s exactly what happened here. You didn’t set out to invent an AI research team—you stumbled into one and then realized how useful the architecture was.