AI has changed how I do design work. I'm trained in the craft, with a career in it since 2004, and I now pair that with Claude every day: more options up front, faster iteration, and working front-end code as part of design review. It has made me much faster.
The craft decisions stay mine. Claude handles the volume and the busywork, so I spend my time judging and refining instead of pushing pixels.
My term for feeding the model my own designs and skills: an established design system, existing UI layouts, and my WWMD review skill. Better fuel means a better first round.
I ask for several directions up front instead of committing early. A wider range means better decisions and fewer late surprises.
I refine by hand in Figma, then feed the changes back to Claude so the next pass builds on my decisions.
Review happens in a running front end, not a static mockup. Stakeholders click through real behavior, and approved code hands off to engineering right away.
Strong first rounds cut time to development and to market.
I came up through traditional design, and that foundation is what makes the AI output good. I know what a strong layout looks like, which option to cut, and when something is almost right. The tools add the speed. The taste behind them has been building since 2004.
I've taught Claude the design system I built for MNTN Select, so it can turn a screen, flow or idea into a working HTML prototype that looks and behaves like the real product. Stakeholders can click through the UX themselves instead of imagining it from static frames.
It extends how I've always worked: I'm a big fan of proof of concept.
Try one: the Budget & Goal prototype ↗With the Figma MCP, Claude reads my designs and components directly and can write back into the file, so design context doesn't get lost in screenshots and copy-paste. I use MCPs, the Model Context Protocol servers that connect an assistant to the tools a team already uses, across my whole workflow.
I've captured my design taste as a review skill I call WWMD (What Would Matt Do). Give it engineer screenshots and the Figma designs, and Claude lists everything that doesn't match, judged against the design system and my own rules. QA gets faster and more consistent, and I spend my time on the judgment calls.
Figma comments and annotated screenshots become clear, well-formed Jira tickets for engineering. Claude drafts, I approve, and it files them through the Atlassian connector. Fewer vague handoffs, fewer rounds of clarification.
I also build prompt-based apps and use ChatGPT alongside Claude. Details are available in conversation.
AI products I designed, and projects made faster with my AI workflow.
Front-end prototypes built with Claude from the Select design system. This is what stakeholders click through in review.
Switch between grid and list, expand a deal, and flip between light and dark.
Mark Content, Demographic, Budget and Timing as learned, in any order, and watch each concept respond.
Add a budget, pick a goal, and switch to the empty and blocked states from the settings icon.
AI earns its place when it removes real work, not when it's added for its own sake.
Be careful not to compromise trust or feel intrusive, especially when a product asks for people's data.
The best use is quiet: collect what can be collected automatically so people do less and get to value faster.
More broadly, I've led 0→1 work at MNTN that includes AI-powered features that improved campaign workflows and surfaced actionable insights. My MNTN Select case study includes AI-assisted media planning.
Read the MNTN Select case study →The design landscape keeps changing. I stay current through continuous learning and build a culture where trying new tools carefully is part of the job.