Multi AI Agent Panel

Problem

UX Designers want everyone to use their product, but when it comes to people with disabilities many designers haven’t had opportunities to gather feedback from the community. With such a wide range of accommodations needed, it can be overwhelming for a UX Designer to read through every regulation from the WCAG and EAA. In addition to that, the business does not provide budget to run early stage UX research with people with disabilities.

How can we get our UX Designers to receive feedback about their product ideas at the earliest stages of the design process to make their designs both compliant and as inclusive as possible?

Concept & Idea

During my course “UI/UX Design for AI Products” I learned about an AI experiment that used multiple AI agents to interact with one another as if they were people in a small village. The results were interesting, with the agents acting very similar to people and carrying out the behaviors they were programed to have. The agents were social, talked about going to another agent’s party, spreading gossip in some cases. This experiment led me to an idea on how to solve the problem for our own UX Designers trying to make inclusive products.

Create multiple AI agents that provide feedback on design concepts from the perspectives of customers with a unique disability. The AI agents would be a digital version of panel of expert consultants you might see in a UX Research panel to provide detailed insights in how they may or may not use a product.

I reached out to a former colleague, now a Stanford Graduate student who specializes in AI, for ways to solve this.

Creating the MVP

Using CrewAI, a tool that allows you to build multiple AI agents that interact with one another, I coded multiple personas in Cursor based on disability gaming consultants I had worked with in the past, along with a UX research agent to summarize the other agents information into a well-written research report.

I then connected CrewAI to Serper to search tens of thousands of documents on the internet related to gaming for people with disabilities and ChatGPT to analyze the information and write the research report.

With a few tweaks and additions to the personas, along with the debugging, the program created a summarized report of the top 20 issues the panel of “gaming disability consultants” would encounter.

Initial Results - Lacking Context & Details

While the reports produced by the AI agents were accurate & helpful to understand what gamers with different types of disabilities experience, they lacked an understanding of the designer’s work to provide useful feedback. In addition, the report was a summary instead of insights that were actionable for the design team.

The positive takeaway was that the information was not “AI Slop” and that we could potentially find other uses for the technology, such as basic desk research that can be time consuming.

Impact

Based on the results of our AI Agents, I pivoted my efforts to create accessibility guidelines that synthesized the WCAG and EAA guidelines while making them contextual for our products (website, mobile app, gaming console, and gaming controllers). This better provided clear and succinct information for our designers to take action on to create inclusive products.