· 5 min read · Design systems, AI

Multibrand design systems with AI, part 1: is it worth it, and where to build it

A palette with blobs of gouache paint in many colors
Image: Jeff Dahl, CC BY-SA 3.0 (resized)

AI can now generate a whole screen in seconds, so is a multibrand design system still worth the work? Often it is. Drawing a button was always the easy bit, the real work is keeping lots of brands consistent and accessible over time. In this first part: should you build one, and where.

What "multibrand" means

One set of components, several brands. A button exists once, and each brand gives it its own colors, type, radius and spacing. Switch the brand and the whole UI changes, without copying any components. Those snippets of information, like button border, button default color, button hover color, text color, text size, button side padding and so on, are called tokens. The best explanation of a token is that it's a design decision.

Is it worth it in the AI era?

AI makes UI cheap. It doesn't make consistency cheap. A model that guesses a blue each time gives you ten blues, a model that is told to use your tokens gives you one. This has improved as LLMs use memory and codebases (which in itself can be a design system, which I mention later). Accessibility is similar, contrast and focus states get checked once in the system and not in every generated screen. And someone still has to decide what changes and who approves it.

So AI is actually an argument for building one, because AI output is only as good as the context you give it. A tokenized design system is very good context. It's the best context. AI works best when it has good examples. If you build components in a design system, AI will be best equipped to do any brand changes and make any upgrades.

Don't do it if you have one brand and no plan to make a business out of rebranding it for multiple customers. It costs real time and money. You don't build it "just in case", you build it when it is part of the business. And one brand with light and dark mode doesn't count, that's just theming.

Where to build it

Figma variables. This is where most people start. Variables (tokens, sort of, it doesn't matter) live in collections, and each collection has modes, for example one mode per brand. Watch the limits. Per Figma's plan comparison, Professional allows up to 10 modes per collection, Organization up to 20, and Enterprise unlimited modes, with extended collections. In an extended collection a parent collection holds the base and each brand overrides only what differs. Figma lists them for Enterprise, Government and K-12 Education plans. A collection can hold 5,000 variables.

I didn't try extended collections. They look interesting, but getting Enterprise Figma and paying that price is a bit absurd to me. Figma's pricing in the top tiers is just not worth it. Count your modes before you commit. If every brand also has light and dark, five brands already need ten modes. The biggest drawback is that Figma is very fiddly with big design systems, it might get laggy, you are limited in modes, and it's hard to work across brands. Changes to already set up systems that are in use are a nightmare.

Tokens Studio for Figma. A plugin that keeps tokens in sets and themes and syncs them to a Git repository. That is the best way to do it. Source of truth lives outside Figma, it loads into Figma. Design tokens are open source. Tokens Studio provides the tooling for easier manipulation. The tooling is extensive, there is support for calculating values. This means you can really create a monster of a design system that changes based on only a few variables. At the time of writing, its pricing page lists a free Start plan with no paid plugin features and one theme group with two options. The Team plan is 39 euro per month billed annually and includes Git sync and multi-dimensional theming, which is really what you want. Multi-dimensional theming means you don't need to create a whole other mode, like in Figma, you can change only what you want. It's more logical, intuitive, and easier to change, manage and support. The drawback is the extra money and a hell of a learning curve. If you want to create a comprehensive design system, get ready for migraines. It's more akin to programming than designing.

Code first. Here the tokens live in the repo and design tools just consume them. Style Dictionary turns token files into CSS, iOS, Android and more, and has first-class DTCG support since v4. If developers own the system, the repo can be the only source of truth. Keep components as finished code, set up theming with AI and you are good to go. The drawback is that you neuter designers. It's hard to communicate with clients, it's hard to do small changes, versioning, design decisions get lost. I think it can work, but you need a really good setup. For big systems with lots of demanding clients, Figma is the standard. You need collaboration with your stakeholders, with clients, it's necessary to communicate the design and the flows. If the setup is such that Figma can build the code and vice versa, at least semi-automatically, then it might work.

Where AI comes in

Figma's MCP server can read everything in Figma. Variables, components, positions on pages, everything. That means you can use AI to build Figma components and you can use Figma components to build them in code. That is a fairly new development: the quality and speed of this have become good enough to be useful.

My recommendation and a short guide

Start with native Figma variables. They have the fewest moving parts. Even if you decide on Tokens Studio later, you will probably still use variables, as they are able to work together.

Next: part 2, structure and variables, then part 3, AI and code.

Sources

Comments