Building UX with AI

A design system should provide repeatable assets and guidance for most scenarios. Sometimes, though, you need a very specific UI, and it’s up to you as a designer or developer to piece it together from what you find in the system.

With AI, it became possible to give builders exactly what they needed on an individual basis. We created a generative search tool that used the Polaris design system to create guidance, patterns, and code based on specific user inputs.

AI is powerful, unknown, and a little scary. So we wanted to get this right. Before starting, we asked ourselves:

• What do we do if AI gives inappropriate responses?

• How do we verify its accuracy?

• What kind of prompts deliver the best results?

When we began testing these questions, we realized that word choice for prompts was absolutely critical. Slight variations could produce wildly different results.