Let’s talk
← All articles

Design systems with AI: structure is the shortcut

AI can generate a screen in seconds. Without a design system behind it, it generates inconsistency just as fast. How I use AI with design systems in iGaming products.

AI tools can turn a prompt into a working interface in seconds. Figma Make, code assistants, UI generators: the first draft is no longer the slow part.

That changes what a design system is for. It used to keep a growing team consistent. Now it also keeps the machines consistent.

Without a system, AI multiplies the mess

Ask an AI tool for a deposit screen and you get a deposit screen. Ask again tomorrow and you get a different one: a new shade of green, a 14px radius instead of 12, a button style nobody approved. Each screen looks fine on its own. Together they look like five products.

In iGaming this costs more than in most industries. Players move between casino, sportsbook and promotions in one session, often across several brands running on one platform. Every inconsistency is a moment of doubt, and doubt next to a deposit button is lost revenue.

AI doesn’t create this problem. It makes it faster.

Tokens are the language AI understands

The part of a design system AI uses best is the least visible one: tokens. Colours, spacing, radii, type sizes and motion, named by role instead of by value. color.action.primary instead of #B6F500. space.300 instead of 24px.

When tokens are named well and live in one place (Figma variables, in sync with code), an AI tool works with intent instead of guessing. “Use the primary action colour” means the same thing in Figma, in code and in a prompt. Switch the brand and every generated screen follows.

On multi-brand platforms this is the whole game: one component library, one token set per brand, and AI producing variants that are on-brand by construction.

Components need rules, not just looks

A component is not only a drawing. It carries decisions: when to use it, which states it has, what it never does. Most libraries keep those decisions in designers’ heads.

AI can’t read heads. So I write the rules next to each component: a short description, its states, the do’s and don’ts, real copy examples. It’s the documentation every team means to write, and now it has a reader that applies it every single time.

The result: when I prompt a new promotion module, the output already uses the right card, the right badge and the right empty state.

Where AI helps me today

  • Exploration. Ten layout directions for a lobby in an hour, built from real components. I keep two and refine them by hand.
  • States and edge cases. Loading, empty, error, long translations, huge win amounts. AI is good at the unglamorous states teams usually forget.
  • Documentation. First drafts of usage notes for each component, which I then edit.
  • Audits. Finding detached instances, hard-coded colours and near-duplicate components across files.
  • Handoff. Prototypes in Figma Make that developers can click through, using the same tokens they will ship.

Where it doesn’t

AI doesn’t know your players. It doesn’t know which bonus mechanic is allowed in which market, why a VIP tier has to feel earned, or when a celebration animation turns into pressure. Responsible gambling, regulation and taste are still design decisions.

It doesn’t decide what goes into the system either. Adding a component is a product decision: does this pattern solve a real, repeated problem? AI will happily generate a thirteenth button variant. Saying no is still the job.

How to start

  1. Name tokens by role. Turn value-based names (lime-500) into semantic ones (action-primary). This alone makes AI output more consistent.
  2. Write one rule per component. Not a manual: one sentence on when to use it, one on when not to.
  3. Keep one source of truth. Figma variables and code tokens must match. If they drift, AI learns the drift.
  4. Review like a lead. Treat AI output as a junior designer’s first draft: fast, useful, never final.

The takeaway

AI makes the first screen cheap. A design system makes the hundredth screen right. Teams that invest in structure now will ship faster with AI; teams that skip it will ship inconsistency faster.

The decisions stay human. The system makes sure they scale.

Got a product in mind?Let’s design it.

Let’s talk