About Me
How I use AI
There's a line in AI-assisted design work, and most of the argument is about where that line is.
On one side of the line, AI frees me up for the work that actually matters: product direction, UX strategy, the real thinking. On the other, it does enough of the work that I've quietly stopped being responsible for it. Like everyone, I had issues with AI, but I got past the skepticism about hallucinations and slop a while ago. I'm still working out where the line falls, project by project. These are a few of the places I've found it.
Making a design system machine readable
Our developers had already taken our Design System documentation site and enabled a plugin to optimize the content for agents. But for my team's workflow, I wanted to take this further. Armed with a directive to reduce dependence on Figma, I wanted to see if I could use AI to build pages with our existing Design System. With Claude, I created a JSON registry as well as skills and documentation that enabled my team and I to build pages with actual UMD Design System front-end components using simple prompts. As the project developed, I created design evaluation skills, QA tests, design brief and document driven workflows and much more.
Almost all of our new project design work now happens in code, using working components in prototypes. The gains are practical. Partners respond better to a prototype that behaves like a real site. And the time we used to spend wiring up pages or faking animations is gone entirely. Changes that meant editing screen by screen are now a single instruction. This hasn't replaced Figma. We can still design parts of a page there and bring them into the prototype when that's the faster route. We've created real efficiencies in our design process.
Turning ideas into artifacts to improve the author experience
Recently I was tasked with putting together requirements to improve the author experience (AX) for a content management system component. The original component AX had been created in a rush and was powerful but complicated. I brought this to my team to get new ideas on approach. A lot of non-technical people were going to use this component; it needed to be foolproof. Most ideas were incremental changes.
My approach was to rethink the process and sketch out how to capture the information in a more simple way, complete with field information and examples. I put my ideas for flow and fields into a document which I brought to my AI tool of choice (Claude) to help me validate it and work through scenarios. Once I landed on a visual format that made sense and could show the logic, we did 6 rounds of refinement before I brought it to the team. Tradeoffs around scope and complexity aside, the work with AI artifacts were a time-saver in communicating requirements and aligning on expected user experience.
User and competitor research
Some of my early work with AI tools included user research. Sifting through and summarizing user feedback and surveys is a no brainer for machine learning. I still needed to bring context and high level review to validate findings, but there's no doubt AI increased efficiencies here.
Work with AI in my process and my team's processes is evolving. There's no denying there are benefits and tradeoffs. I'll continue to explore efficiencies to benefit my workflows while maintaining AI's use as a tool and not a replacement.