The Virtual Roundtable: Multi-Agent Personas Simulating the Dynamics of Human Brainstorming

Researchers present a multi-agent AI system that simulates human brainstorming through diverse AI personas engaging in structured roundtable discussions. The architecture uses divergent and convergent thinking phases to generate and evaluate ideas while minimizing groupthink, demonstrated through a case study on AI smart glasses product concepts.

This research addresses a fundamental challenge in AI-assisted product development: moving beyond individual AI capabilities to simulate the creative dynamics of diverse human teams. Traditional brainstorming suffers from well-documented cognitive biases—groupthink, echo chambers, and dominant personalities drowning out minority viewpoints. The multi-agent approach directly counters these failure modes by embedding diverse personas with distinct perspectives into a facilitated discussion framework.

The system's two-phase methodology mirrors proven brainstorming techniques. Divergent thinking generates volume and variety without premature judgment, while convergent thinking applies critical evaluation. The inclusion of private thoughts alongside public commentary creates psychological safety, a key ingredient missing from many collaborative AI systems. Quota mechanisms ensure balanced participation, preventing any single perspective from dominating outcomes.

For product teams and development organizations, this represents a scalable alternative to assembling diverse human brainstorming sessions, which require significant time and coordination overhead. The approach captures idea lineage—understanding how concepts evolve and cross-pollinate—providing strategic insight into creative processes typically lost in unstructured discussions. This auditability matters for organizations needing to justify innovation decisions.

The smart glasses case study validates the method's practical application, though real-world impact depends on whether generated ideas translate to viable products. The broader implication extends beyond product development: as organizations increasingly rely on AI for strategic decisions, incorporating structured diversity into AI-driven processes becomes critical infrastructure. Teams should monitor whether this approach scales to complex domains requiring deep domain expertise, where persona authenticity matters more than novelty.

• →Multi-agent systems can replicate beneficial aspects of human brainstorming while reducing cognitive biases and groupthink.
• →Structured frameworks with divergent and convergent phases produce higher-quality ideas than unmoderated AI generation.
• →Idea lineage tracking provides strategic insights into how concepts evolve and recombine during collaborative processes.
• →Private thoughts paired with public commentary creates psychological safety that encourages diverse perspective expression.
• →The approach scales brainstorming capacity for product teams without proportional increases in coordination overhead.