SKU PROJ-010 · Release 2026
In most teams, design questions come in faster than designers can answer them. The Brand Assistant opens the art direction to everyone: anyone can find the right rule, start a mockup or an asset, and the designer approves what goes out.
A product manager wonders which colour to use for an alert. Marketing needs an on-brand visual by tomorrow. A developer hesitates between two components. All these questions land on the same design team, which is almost always understaffed. Designers spend their days answering instead of designing, everyone else waits or improvises, and the brand gets diluted one screen at a time.
Turn the art direction into a service anyone can use, at any time. An assistant that answers design questions, audits a screen, creates or edits a mockup. One condition: never step outside the brand rules.
Art direction never lives in one place. Tokens and components sit in the design system. Imagery, illustration and iconography, in marketing content. The way to write, in the tone of voice. Step one was to gather everything, then check and date every source. A recent capture beats an old one, and an official document beats a third-party record. When two sources disagree, the more reliable one wins and the gap is logged.
A language model reads structure well. The art direction was split into short cards, one rule per card, all built on the same template: the rule as one testable sentence, its intent, when to apply it, what to do and avoid, its source. Each card gets an ID that will never change, and complete metadata: type, screens concerned, expected emotional level, status, confidence level. It is the longest and least visible part of the work.
The roughly 200 cards live in a single rule base. Every component in the Figma library carries its card's ID, and every card points to its component. The same base serves two audiences: the team, who filter and find their answer in under a minute, and the assistant, which is not allowed to draw from anywhere else.
The model was constrained as rigorously as an interface component. Its rules of conduct sit in the same base as the brand, with their own IDs. It only answers from the cards and always cites the one it applies. With no rule, it answers "not documented" and suggests the card to create. It writes like a busy art director: answer first, a few sentences, sources on the last line. It only uses existing components and colours, and never adds anything to the library on its own. Every statement shows its confidence level, and anything resting on an assumption stays a suggestion, never an imposed correction.
Using the assistant comes down to four uses, each with its own deliverable.
The assistant acts, then shows, and nothing is ever overwritten. Every deliverable lands in a versioned Figma section that puts side by side a changelog (the request, the rules applied, the date), the original mockup, cloned and untouched, and the new version.
Every exchange is logged, from the question to the final decision. The assistant states what it is doing in its first sentence, so it can be stopped, and only pauses on its own when a decision commits the brand. Everything it produces stays a draft until a designer reviews it, and sensitive topics such as pricing, legal or accessibility require a closer review.
The demonstrator was first built on the art direction of a major consumer brand. It now runs white label: the assistant, its rules of conduct and its 4 flows stay the same, only the rule base changes. Plugging in another brand means giving it another rule base, structured the same way. A rule base built on Yeita's design system is already in place.
DS Setup is a tool that draws on Figma and GitHub to make them the design system's source of truth: components, their properties and their tokens are read where they actually live, on the design side and the code side. That base feeds the rule base, instead of documentation copied by hand that quickly goes stale.
A working demonstrator across the 4 flows, traceable at every step, not yet in production. Next measure: the time saved by the team, and the share of requests that no longer need to wait for a designer.
Connect the assistant to DS Setup continuously. It could then compare Figma and code component by component (same names, same properties, same tokens) and flag every gap before it reaches production.