One product design cycle, before and after an AI-ready design system and a multi-agent workflow.
Global pharmaceutical company, clinical supply
- Frame
- Give AI context
- Prototype
- Check
- Hand off
- Show the impact
A control tower designed with humans and agents together
The people who keep medicines moving to clinical trials were juggling several separate supply chain tools. The goal was one platform and one view. I led the engagement and the UX team. We started with ordinary UX deliverables and ended with something bigger: a design system built in React that both people and AI agents could use, and a new way of working that cut a design cycle from four weeks to just over one.
Read the full story
What we built
A unified design system with components, tokens, and written rules that agents could read, so creating and checking designs became repeatable. A workflow where specialized AI agents handled parts of the work alongside our designers. A direct path from prototype into the client's React code, with new components flowing back into Figma so design and engineering worked from one shared library. Plus full-day workshops, executive materials, and the screen naming conventions the client's training team now uses.
What speed taught me
- Structure keeps AI honest. Where the design system was thin, AI misread intent or made things up. Where approved components and clear rules existed, the output stayed accurate.
- The round trip still has friction. Early tool limits pushed us straight into React. That kept us moving, but bringing new components back into Figma still takes extra steps.
- Speed exposes the human bottlenecks. Once design got fast, the slow parts were data access, decisions, and expert time. AI can't sprint through those.
- Evaluation never ends. The tools change monthly. I treat tool choices as ongoing experiments and judge every engagement by its outcomes.
The team: Inez Yuen, Tharun, Pravin, Chetan, Yilmaz, Tony Knaff, and Umang. This story is theirs as much as mine.