Starting to work with GenAI as a UX designer feels awkward at best. For the next few months, I have the opportunity to run some experiments and I’ve been trying to use it as a chance to see how this new tool might affect my team's ideation process.

• How do I help users overcome the same awkwardness I'm feeling? What unique value can overcome the hesitation a user might be experiencing?

• How do I even use this thing? The blank canvas problem is real and from my experience, I find open input a barrier to getting started.

• Why would anyone trust the output? The “creativity” that makes LLMs unique also makes the output unreliable.

I found several resources aimed specifically at UX designers as we begin to experiment with this new material.

Design Patterns for AI Interfaces by Vitaly Friedman was a great kickstart hyper-focused on UX designers. While many resources are either general intros to LLMs or targeted at design process utilities, this course details specific AI UX patterns that directly shape the user experience. I loved a healthy dose of skepticism mixed with real-world examples.

Shape of AI by Emily Campbell is one of the many examples referenced in the above course and the most essential for UX designers. The site establishes a robust vocabulary of patterns that map fairly well to an end-to-end-user experience journey. I appreciated the focus on several patterns that promote a more responsible approach given the risk of misuse.

Sentient Design by Josh Clark was a great resource that gets more technical into how LLMs fit within the larger landscape of AI and machine learning tools. I appreciated the mix of optimism and candor about the potential missteps when we design experiences with this technology. It was exciting to see examples of interfaces that could leverage user input and context on the fly to generate custom UI.

Some big takeaways from these resources targeted specifically at UX designers included

• Start by identifying appropriate use cases and disclose to users any risks and shortcomings of using the tool before they get started to align expectations.

• Experiment with suggestions and structured templates to provide training wheels and allow users to form their prompts into something your tool can leverage into more valuable outputs.

• Whenever possible, provide citations annotating sources used in creating the output, watermarks that identify the output as generative, as well as footprints allowing users to see how the system created the output for easier validation.