TeamFusion: Supporting Open-ended Teamwork with Multi-Agent Systems
New framework uses AI agents to represent team members, surfacing disagreements to build stronger consensus.
A research team led by Jiale Liu has introduced TeamFusion, a novel multi-agent AI framework designed to tackle the complex challenge of open-ended teamwork. Unlike traditional aggregation methods that simply average opinions—often suppressing minority perspectives—TeamFusion creates a personalized proxy agent for each team member, conditioned on their expressed preferences. These AI proxies then engage in a structured, multi-round discussion to explicitly surface areas of agreement and disagreement, moving beyond simple voting to resolve underlying conflicts.
The system operates through a three-phase cycle: instantiation of member proxies, structured discussion, and synthesis of consensus-oriented deliverables, which then feed back into further refinement. The researchers evaluated TeamFusion on two distinct teamwork tasks, measuring both how well individual views were represented in final decisions and the consensual strength of the deliverables. Results showed it consistently outperformed direct aggregation baselines across various metrics and team configurations, demonstrating its effectiveness in producing more robust and inclusive collaborative outcomes.
- Creates personalized AI proxy agents for each team member based on their preferences
- Uses structured multi-agent discussion to explicitly surface and resolve disagreements, not just aggregate votes
- Outperformed direct aggregation baselines across two evaluation tasks and multiple team configurations
Why It Matters
Enables more effective remote collaboration and complex decision-making by preserving diverse perspectives in AI-assisted teamwork.