Build your AI knowledge
Posted: Sun May 25, 2025 7:17 am
Learn about natural language processing and large language models. Discover how on Trailhead, the free online learning platform from Salesforce.
Start learning
+1,000 points
Trail
Explore Generative AI Tools
AI transforms the future of UI (user interface) design dramatically. Instead of apps, you can tell AI your goal, and it can accomplish it for you – no clicking or app learning needed. AI is so incredible because it has the potential to shorten the distance between the goal and the accomplishment.
What’s deterministic versus probabilistic design?
How does AI affect the future of UI design? In traditional product design, we create the interface based on an understanding of a predictable (deterministic) set of data on which the app operates. Designing an app to track your mood? You’ve got a static data model plus a deterministic interface to read and write data into the database. Designing for AI, however, means working with data that is often changing and therefore probabilistic. This shifts the way we design our user interfaces.
At Slack and Salesforce, we’ve been designing for a number of AI features, and here are a few insights from what we’ve learned so far.
Tip 1: Embrace probability
When we design for AI, the data can be influenced by the large afghanistan phone number list language model (LLM) we use. It also can be impacted by the quality of the prompts we give the LLM. Because the data isn’t pre-determined, the results can vary. In these image examples from Midjourney, both are about the same topic. But different prompts give us radically different results.
"Happy designers" prompt yields a shot of young designers smiling at the camera and wearing typical button-up shirts. A second image shows two people dressed in what appears to be costumes and hats.
The same topic with different prompts yields different results.
So when you’re designing a product that incorporates AI, you need to:
Determine the rough outline of the interface where the result will live.
Run the prompt to get a feel for the data output.
Use the data output to validate the outline of the interface.
Refine the prompt and the interface outline according to the results.
This is an iterative process. As you refine the interface and the output into something more concrete, you may find that revised prompts provide much better data, and that this data requires different expressions in the UI.
Start learning
+1,000 points
Trail
Explore Generative AI Tools
AI transforms the future of UI (user interface) design dramatically. Instead of apps, you can tell AI your goal, and it can accomplish it for you – no clicking or app learning needed. AI is so incredible because it has the potential to shorten the distance between the goal and the accomplishment.
What’s deterministic versus probabilistic design?
How does AI affect the future of UI design? In traditional product design, we create the interface based on an understanding of a predictable (deterministic) set of data on which the app operates. Designing an app to track your mood? You’ve got a static data model plus a deterministic interface to read and write data into the database. Designing for AI, however, means working with data that is often changing and therefore probabilistic. This shifts the way we design our user interfaces.
At Slack and Salesforce, we’ve been designing for a number of AI features, and here are a few insights from what we’ve learned so far.
Tip 1: Embrace probability
When we design for AI, the data can be influenced by the large afghanistan phone number list language model (LLM) we use. It also can be impacted by the quality of the prompts we give the LLM. Because the data isn’t pre-determined, the results can vary. In these image examples from Midjourney, both are about the same topic. But different prompts give us radically different results.
"Happy designers" prompt yields a shot of young designers smiling at the camera and wearing typical button-up shirts. A second image shows two people dressed in what appears to be costumes and hats.
The same topic with different prompts yields different results.
So when you’re designing a product that incorporates AI, you need to:
Determine the rough outline of the interface where the result will live.
Run the prompt to get a feel for the data output.
Use the data output to validate the outline of the interface.
Refine the prompt and the interface outline according to the results.
This is an iterative process. As you refine the interface and the output into something more concrete, you may find that revised prompts provide much better data, and that this data requires different expressions in the UI.