Our product is a nationally deployed clinical EMR with a UI that hasn't meaningfully changed in about 30 years. It's in use across many client jurisdictions, each with its own set of providers. That sounds fine internally, the system works and clients are used to it, until you compare screenshots against the competition. Larger companies solve this with dedicated frontend and UI/UX headcount. We don't have that team. Our strengths are support, custom requests, and responsive customer service, none of which show up the moment someone opens the software for the first time.
Why nobody had fixed it
The design was far enough behind that even modest effort produced a visible jump, and still, technical debt and feature work kept crowding it out. The deeper issue: most of the team is functionality-oriented. They can ship a feature that works correctly, a real skill, but making it look like it belongs in 2026 needs React, CSS, and HTML fluency plus an eye for which data deserves the user's attention. Not everyone has both halves.
The eye I had, the skills I didn't
I worked as a frontend developer before this job. That gave me the eye: I know what a clean interface looks like and how to surface the data that matters instead of burying it. What I'd lost was the day-to-day fluency to build it myself, fast, at the level the product needed. I've designed dev dashboards and other tools I think look good, and AI has been what bridges having a design idea and presenting it working and styled. That gap used to be my bottleneck.
I used AI to architect the implementation: reusable styles instead of one-off overrides, a real CSS architecture instead of patchwork, built so the next update doesn't start from zero. I folded ADA compliance into the same pass using the rules behind the compliance agent, since accessibility was a rating we'd have to chase anyway and doing it here meant not doing it twice.
The demo
I demoed the updates in our UI/UX meeting. The senior dev said it should go out immediately, if contract acceptance and the approval process weren't in the way. That's the bottleneck now: the process around getting it in front of clients, not whether the work holds up.
This is the same shape as the MCP server work: a gap with no dedicated headcount, and AI closing enough of the distance that I can be the one who does it, on an eye for design I had and skills I didn't.
Update: a second senior dev saw the demo and asked for it on our internal site. See the follow-up on rebuilding that frontend in three days.