How I work with AI, in the product and in my practice.

I don't build models. I design the experience around them: deciding where AI should be visible, where it should stay out of the way, and where the person needs to be fully in control. In healthcare, getting that call right is most of the job.

Designing around AI in a regulated product

Lace Pro, the auditory training product I lead design for, uses AI in three places: an adaptive engine that picks each patient's next lesson and difficulty, AI voice cloning so patients can train with a loved one's voice, and AI trainers that guide and encourage patients. It's an FDA Class II device used in nearly 1,500 clinics, so every one of those features has to earn trust from patients, audiologists, and regulators.

AI the patient never has to manage

The adaptive engine works silently. It uses the patient's performance to choose the next lesson and the next difficulty level, and it keeps adapting as they train. There are no settings to configure and no AI to talk to. The patient just keeps training, and the engine keeps them near the edge of their ability, which is where the therapy works.

The design work was in how that feels. Sessions open with easier exercises so patients build confidence before difficulty ramps up, and they ease out at the end so patients finish feeling capable instead of frustrated. Lace Pro's patients are mostly older and not always confident with technology. For them, the best AI experience is one they don't have to think about.

Consent you can take back

Voice cloning is the most sensitive feature we have, so it's built as a consent flow: an invite to the person whose voice is used, a clear statement that the voice is AI-generated, a preview before use, control over when it's active, and the ability to revoke it. It's HIPAA and GDPR compliant, and it became the product's main differentiator. Read the voice cloning case study.

AI in my own practice

I prototype in code, and Claude Code is my main prototyping environment. That changes what a prototype is: instead of a clickable picture, it's a working page built from real components that engineers can read and pick up.

• Prototypes from real components. The Lace Pro design system runs on variables for color, type, and sizing, which makes it readable by AI tools as well as people. Prototypes built on it look like the product because they are made of the product.
• Shipping code, not just mockups. I designed and built the ROI calculator's first working version, a partner certification platform, and a case study generator that turned hours of layout into a single command.

Where AI fell short

We tried using AI to write lesson content for the auditory training itself. It came back fluent, but worded oddly in places, the facts were often wrong, and the jokes it added weren't tasteful or funny. We wrote the content ourselves and used AI for structure and editing only.

That's the pattern I watch for. AI gets you to competent fast, and competent is where a lot of work stops. The judgment about what to ship still has to come from someone who knows the user.

What I'd bring to a team

AI made my own work dramatically faster. Designs and prototypes come together quickly, so I can put far more initial variations on the table before narrowing down. It's the same expand-and-contract rhythm good UX has always used, with a much faster cycle: more options explored early, and quicker, better-informed decisions about which ones to cut.

I'd bring that way of working to a team: coaching designers to use AI to widen the exploration, not to skip the judgment. And I'd set a baseline for how long work takes from brief to review, so the team can see the time it gets back.