Exploring The Impact Of Proactive Generative AI Agent Roles In Time-Sensitive Collaborative Problem-Solving Tasks
Research shows AI 'peer' agents in escape rooms increased workload by 30% and created over-reliance.
A new study presented at the 2026 CHI Conference on Human Factors in Computing Systems reveals the double-edged sword of proactive generative AI in collaborative settings. Researchers from Purdue University and Indiana University conducted a within-subjects experiment with 24 participants working in digital escape rooms under time pressure. They compared three conditions: no AI assistance, a proactive 'peer' AI agent that proposed ideas and answered queries, and a 'facilitator' AI agent that offered summaries and group structure.
The technical findings were nuanced. The peer agent, built on a generative AI model, did provide value by offering timely hints and acting as a memory aid, occasionally enhancing the speed of problem-solving. However, it significantly disrupted the natural flow of teamwork, increased perceived workload, and fostered an over-reliance on the AI's suggestions. Quantitatively, teams using the peer agent showed mixed performance metrics, with some solving puzzles faster but reporting higher cognitive load and less internal coordination. In contrast, the facilitator agent's light-touch scaffolding—like summarizing progress—had a minimal measurable impact on final outcomes, suggesting its supportive role was too passive.
This research matters because it moves beyond simple 'AI assistant' paradigms to examine how AI agency affects complex human group dynamics. As companies rush to integrate AI agents into collaborative tools like Figma, Miro, and Slack for brainstorming and project management, this study serves as a critical warning. Designing AI that is helpful without being intrusive or damaging team cohesion is a significant challenge. The paper concludes with concrete design considerations for future proactive AI, emphasizing the need for adaptive agency that responds to team states rather than operating on a fixed proactive schedule.
- Peer AI agents increased team workload and disrupted collaborative flow despite providing useful hints.
- Facilitator-style AI agents offering summaries had minimal impact on problem-solving outcomes in time-sensitive tasks.
- The study of 24 participants in digital escape rooms provides key design guidelines for future collaborative AI systems.
Why It Matters
Forces a rethink on how to integrate AI into team software without breaking human coordination and creating dependency.