A foundation the team builds from now, and agents next.
Unica NEXT is a building-management platform. Its design system had drifted: intent in people's heads, the library out of use, design and code pointing at different things. Phase 1 rebuilt the foundation as one connected source, so the team builds from the system now, and AI can build from it next.
The problem
A lot of the system lived in people's heads
The team had a real component library, but it had drifted out of active use. A lot of what design handed to development was no longer connected to it: tokens were not aligned, the documentation no longer worked as the living source of truth, and what left the design space was only partly systemised. Some of it was true reusable components. The rest was interpreted, rebuilt, or redrawn in code, depending on who was working on it.
That works as long as the right people carry the context. It does not scale to a shared system, and it definitely does not scale to an LLM, which would do exactly what a new person does where the system is unclear: start guessing.
What left design was only partly systemised. The rest was reinterpreted in code or carried in people's heads, fine until you need a shared system, or an LLM.
My role
What I owned
I led the foundation phase as freelance design systems lead, working solo. The brief was the first move: get the foundation back on track. I set the token architecture, paired the components between Figma and React one by one, stood up the living docs site, and built the custom plugin that keeps Figma and the repo on the same tokens.
I also put the rails in place: the repo as the source of truth, Code Connect linking design to code, and automated audits that flag drift instead of letting it hide. Not a big transformation programme, a sequence of small connected moves the team can build on.
The approach
The first step isn't AI. It's the foundation.
Unica builds smart buildings, where nothing works unless everyone can see what connects to what. The same is now true inside a product team. When intent is vague, patterns live in heads, and design drifts from code, you do not get acceleration. You get noise. So it is Unica's own logic, reflected back into how they build product.
Phase 1 made the system explicit. Tokens became named decisions instead of hex scattered across files, so design can change a value, a theme, or a brand decision without a developer ticket. Each component is paired one to one between Figma and React. A living docs site made the system explorable in the browser. And a custom plugin keeps Figma and the repo on the same tokens, with the repo holding the truth.
An LLM does not remove ambiguity. It runs into it faster. Get the foundation reliable first, and then the team builds from it now, and agents can build from it next.
One source. Tokens flow to Figma and code, components stay paired, audits catch drift, readable by the team now and by agents next.
The outcome
The foundation is in place
The foundation phase is done. The system is reconnected: design tokens for light and dark, more than forty components with live React docs being paired one by one between Figma and code, a custom Figma-to-repo sync plugin, and automated audits that surface every mismatch instead of hiding it. Around a hundred small ones are logged for cleanup, not buried. The team builds from one source now, instead of memory and interpretation.
A design system needs ongoing care, whatever the team or product size, and that upkeep is not owned by design or development yet. The phases ahead, making the patterns explorable and bringing AI into the team's way of working, are where this is heading. We are in talks to continue, in a more agile way, toward that agent-ready point. The foundation is reliable enough to build the rest on.
Living case, updated as each phase is signed off.
Questions
Is the Unica design system AI-readable yet?
The foundation is in place: tokens, Figma and React components being paired one by one, a living docs site, and the decisions behind them written down, all from one source. That foundation is the prerequisite. Making it genuinely agent-ready, with guardrails, skills, and AI workflows, is the next phase, now in motion.
Why fix the foundation before adding AI?
An LLM does not remove ambiguity, it runs into it faster. If patterns live in people's heads and design drifts from code, AI adds noise before it adds value. Reconnect the foundation first, then AI has something reliable to work from.
What does one source of truth mean here?
Tokens are authored once in the repo and flow to Figma through a custom sync plugin and to code. Components are paired Figma-to-React via Code Connect, and automated audits flag any drift. Design, code, and docs all point at the same system.
Your design system has two readers now. Most are built for one.
That's the work I do. Let's talk.
Get in touch


