Industry
Fintech • Payments
Client
Computop • Nexi
AI-Assisted Design System: Extending design coverage beyond a one-person team
Situation
One designer, more pressure, less time
Following Computop's acquisition by Nexi, the team was restructured for efficiency. As the sole designer supporting multiple PMs, I was consistently in reactive mode, producing quick mockups in Figma to unblock tasks rather than building the design infrastructure the product needed. There was no design system, developer collaboration was historically waterfall-based, and requests arrived faster than I could address them strategically.
Task
Create time that doesn't exist
To establish a design system, I first needed to reduce the time I spent responding to routine PM requests. But I couldn't reduce that time without something to replace me. The problem was circular. I needed to break it from a different angle.
Action
Building a system that extends my reach
I started by defining design tokens for the payment page (the most contained and consistent surface in our product) and structuring them as a repository of markdown files optimised for AI consumption. I published the repository on GitHub and connected it to Claude's design tooling, which can read the repository and use it as a source of truth for generating and modifying UI. The practical output is a workflow where a PM can generate a payment page layout, then interact with specific elements directly — pointing to a field and requesting changes in natural language. The AI references my design system and applies changes consistently, without me in the loop for every iteration. The intended next step is connecting this to Figma directly, so that PM-generated layouts can be pushed into Figma as a starting point. I can then jump in, refine the design manually, and maintain quality control without being the bottleneck for the initial structure. The longer-term architecture includes a feedback loop with the Nets design team — Nexi's larger design function in Denmark. If their UX research identifies improvements to a component, I update the repository, and the AI inherits that knowledge. One designer, one repository, many surfaces.
