Que signifie vraiment la préparation à l'IA d'un design system ?

Three design experts told me how AI is changing the way we build and maintain Design Systems. And led me to think that “the future” is already happening now.

You’ve probably read, or even written, one of those LinkedIn posts about someone building a complete Design System in 20 minutes with Claude. Or in 15 minutes with Gemini. Or… you get the idea.

Those posts made me curious. I had never built a full Design System, although I had set up the basic elements of one before I really dug into AI, and it was as interesting as it was painful. So the promise of building a full one in minutes was appealing.

My mistake? Confusing a UI kit with a Design System, and believing the whole process could really be done in minutes.

At UXDX Berlin last May, two talks about Design Systems made me want to know more about how AI has changed them in the day-to-day of product teams. That curiosity led me to interview three design experts, and to get to the essential part of what a Design System actually is: Marianne Ashton-Booth, Head of Product Design at ITV; David Sward, Chief Design Officer at Siemens; and Nicolas Chatelain, freelance Senior Designer and specialized in Design System and Ops.

Three experts, three different approaches. Nicolas Chatelain bets on a frame that bends without breaking, flexible with teams and ready for AI. David Sward bets on speed with guardrails, with AI as a tool and a human in the loop. Marianne Ashton-Booth bets on people and process, treating the Design System like a product.

The last two had both just given a talk at UXDX when I talked with them, so what follows draws on two different moments with each of them: what they said on stage, and what they told me one-on-one afterwards.

1. AI automates the easy part, but the hard part still belongs to humans

One-on-one after her talk, Marianne Ashton-Booth said something that felt particularly on point to me: “Anybody can spin up prototypes in minutes, and that’s great. But then that gap between what’s actually feasible, what’s actually close to the brand, or close to the actual components that we’ve got, can be quite distant.”

Nicolas Chatelain, whom I interviewed in July, put the same idea in other words: the UI is the easy part of a Design System. “The hardest part, the one that makes a DS work or not, is governance and adoption.”

Neither of them sees the components as the problem; Nicolas even says anyone could, indeed, build a kit while we were talking. What matters is consistency across teams, and across a whole company, and that is the human part of a Design System. A Zeroheight report on the state of Design Systems, published in early 2026, points the same way: 59% of practitioners credit communication and community for their system’s adoption.

On stage, Marianne Ashton-Booth had already named what breaks instead. When a design system fails, she said, “it’s almost never just [the design system’s core operations]. The components are usually fine.” What breaks are the invisible layers underneath: trust, communication, prioritisation, relationships, and shared belief.

2. The bottleneck has moved from production speed to review

“My biggest problem is trying to slow people down with AI,” David Sward told me. Indeed, AI has sped up a lot of processes, all the way to delivery.

Here, he does not mean slowing his team down so that they get used to AI, but keeping them inside the security and compliance rules of a large enterprise. With hundreds of products relying on his system, a mistake would take several hundred million euros to fix. His operating principle: “don’t break things.”

But the fact remains: productivity has risen. And once speed is there, the bottleneck shifts from production to the people who review, validate and keep everything coherent. Nicolas Chatelain and Marianne Ashton-Booth ran into this too.

During Nicolas Chatelain’s mission for an international group, for example, AI-boosted product teams started to outpace the central Design System team, whose original strategy was to lock everything down. His verdict:

“With AI, it’s a total rollback. We have to become much more flexible, because tomorrow people could create a kit in five minutes. If we become useless, they’ll find an even easier way to bypass us.”

Figma’s slots, which let teams put their own content inside a standardised component, are one way to loosen the frame.

At ITV, Marianne Ashton-Booth’s team takes another route and relies on “design system reps” in both design and engineering: “They tend to be the go-to people that have a little bit more prominence and involvement in things like final decisions. It just means that things can keep moving rather than just becoming a bit of a bottleneck.” On stage, she added that rather than treating AI as a separate topic, ITV asked the right question: “What does that mean for the design system? How should our processes evolve?”

These three approaches suggest that there is no single way to handle the bottleneck: it depends on the context, the level of risk, and how far AI has advanced, and been accepted, in the organisation. That is precisely where humans remain useful: understanding at what moment, and to what extent, everyone who holds decision-making power over the Design System should step in, and when they should let go.

3. The Design System is becoming something AI reads

This is the part that sounds most like “the future”, but is actually already the present: making the system readable for machines.

Nicolas Chatelain’s team started watching tools like V0 and Lovable in the second half of 2025, and concluded it was too early to adopt them. They chose to prepare the system instead: tokens migrated to the DTCG (Design Tokens Community Group) standard, a shared format for the values of a system, like colours, spacing or typography, so that every tool, and every AI agent, reads them the same way.

An MCP server (Model Context Protocol) lets an AI agent query the Design System directly, for instance to find the right component or token, instead of guessing. In a field study published in July 2026, Kaelig Deloumeau-Prigent looked at 21 open-source Design Systems: 20 of them offer an MCP server (official or community-built), but only 3 have set up Figma Code Connect, which ties a Figma component to its real code counterpart. Plenty of systems let AI read them, far fewer tie their components back to design.

David Sward’s team uses AI on the other side of the process: a large language model (LLM) compares what teams have implemented with what the Design System is supposed to be, and, in his words, “systematically tells you what to fix.” In his own Design System, AI stays “a tool just like anything else” (an engine harmonises thousands of icons, for instance), and his broader principle holds: “You can’t remove a human from the loop and then expect them to step back in to operate the system if they don’t understand the state the system is in.”

Marianne Ashton-Booth, on her end, is adamant: “There might be parts of [a design system] that become more AI-generated or automated. But a design system isn’t just that part, you know, it’s the people and it’s the process as well. And that’s the bit that I think we’ll still retain.”

Where I think we’re heading: code as the source of truth

These insights led me to believe this is where Design Systems are heading: the code, rather than the design file, is becoming their source of truth. We might already be there. Designers who use AI seem to open Figma less and less (that’s my case), and in teams where shipping and iterating beat perfecting details, the blurring of roles (I talked about it here) doesn’t help.

Nicolas Chatelain said it: “Today, the source of truth has to be the code. Absolutely.” Working in the code removes a layer of abstraction between AI and production, and “tomorrow, when this documentation lives directly in the code, we won’t need Figma anymore.” He admits he may be “completely wrong”, and still finds Figma useful to explore and fine-tune. And I agree with him that Figma isn’t dead. But for me its place in the process has definitely changed.

Concretely, it means that whatever is merged into the code becomes the standard, and design files, documentation and AI agents all read from it. Every review then becomes a governance decision. But as Tony Ward wrote in his recent article “Code as the Source of Truth”, the key is to keep “a subject-matter expert in the loop at every step of the way.” Which leaves the real question: in this new process, who gets to say what goes into the truth?

As for me, meeting these three experts showed me there is no single answer to that question. Each built theirs from their own environment, and in that richness I found keys to meet my own challenges. I hope you will too.

Thanks to UXDX for providing access to the event, and to David Sward, Marianne Ashton-Booth and Nicolas Chatelain for their time.

All of Nicolas Chatelain’s quotes were translated from French.

What AI has changed in Design Systems, and what is still up to us was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.