Personality Panel

Five AI product managers with different personalities debate your roadmap, then converge.

Give the panel a product and a list of features. Five AI product managers, each built on one of the Big Five personality traits, rank the features on their own, read each other's reasoning, revise, and converge on a priority order. Disagreement is the point: it surfaces blind spots a single perspective misses.

The agents

The Visionary (high Openness): Drawn to novel, differentiating bets and features that open new possibilities for users.
The Executor (high Conscientiousness): Prioritizes reliability, compliance, quality, and work the team can actually deliver well.
The Champion (high Extraversion): Favors features that drive engagement, sharing, visibility, and a great story for customers and sales.
The Advocate (high Agreeableness): Champions user well-being, accessibility, inclusion, and what keeps customers and teams aligned.
The Risk Spotter (high Neuroticism): Focuses on security, failure modes, customer frustration, and what happens when things break.

How this was built

Prioritization meetings tend to reflect whoever is most confident in the room. I built the first version of this as a multi-agent notebook with Strands Agents to show product managers a different pattern: deliberately diverse perspectives, independent first opinions, then a structured round where everyone sees the others' reasoning before converging.

Each panelist is built on one of the Big Five personality traits, the most widely used model of personality in psychology. The panelists run in parallel, and the final ranking is a plain ranked-vote count of their second-round positions, so the math is transparent. The AI's job is the reasoning: why each panelist ranked as they did, and what the disagreements reveal.

Choosing the model mattered more than usual. In side-by-side tests, most models let every panelist converge on the same "safe" answer from the first round, which defeats the purpose. DeepSeek V3.2 kept the personalities distinct in round one and still let them genuinely move in round two.

A demo of multi-agent reasoning for product managers, October 2026. Runs on DeepSeek V3.2 in Amazon Bedrock.

This demo is interactive and needs JavaScript.