How to Use AI to Build and Maintain a Design System

A design system is two things stacked on top of each other. A set of decisions about how your product looks and behaves, and a mountain of documentation and consistency work.

AI is no help with the first one. It is very good at the second, which is also the part that kills most design systems, because nobody wants to do it in year two.

Naming and organising tokens

Ask it to propose a consistent naming scheme for colour, spacing and type, and to flag where your current names overlap or are ambiguous.

The flagging is the valuable half. Ambiguous token names are invisible to the person who wrote them and obvious to everyone else.

Drafting component documentation

Give it a component and its props and have it draft usage guidance, the cases where it does not apply, and the accessibility notes.

You end up editing for accuracy instead of writing from a blank page, and editing is roughly ten times faster.

Auditing for drift

Feed it your styles or component list and ask where things have drifted. Near duplicate colours. One off spacing values. Two components that should have been one.

Every system accumulates this. Nobody notices until it is bad enough to be a project.

Checking for missing states

For a new component, ask what states and variants you might be missing. Loading, empty, error, disabled.

It is a completeness check that takes a minute and catches the thing you would otherwise find in QA.

Questions I get asked

Should it decide token values? No. The values are your brand. Let it help with structure and naming, which are engineering problems, and keep the taste.

Can it keep the system consistent over time? It is a good auditing assistant. Somebody human still has to own the direction, and a system without an owner drifts no matter what is auditing it.