What an AI agent can actually do with a design system

13/08/2026 — admin@byqreal.test
What an AI agent can actually do with a design system

A model asked to build UI from nothing invents plausible but inconsistent interfaces. Given a documented set of components, tokens and rules, it assembles from what exists instead. The determining factor is whether the design system is machine-readable, not how capable the model is.

Ask a model to build a pricing page with no further context and you get something competent and unfamiliar. Reasonable spacing, invented colours, a card component that resembles yours without matching it. It is plausible, and it is not your product.

Give the same model your components, your tokens and your rules, and the output changes character entirely. It stops designing and starts assembling.

Invention is the default, and it is the problem

A model with no constraints fills gaps with what it has seen most: generic spacing scales, common colours, the average shape of a card. Every generation does this independently, so nothing matches anything else.

Constraints are not a limitation here. They are the entire value.

Readable means readable by a program

A design system that exists as a Figma library and a well-designed website is documentation for people. For assembly you need the inventory in text: component names, the props they take, the tokens that exist, what may be composed with what.

Names are instructions

This is the part that surprises teams. A component called Card2 tells a model nothing, so it guesses from context. A component called PricingTierCard tells it what the component is for, and it gets used in the right place.

The same applies to tokens. --blue-600 describes a colour. --surface-raised describes a job, and a model choosing between jobs makes better choices than one choosing between hues.

Rules beat examples for constraints

Examples teach shape well and boundaries poorly. If a spacing scale has six values, saying so is far more reliable than hoping the model infers it from three samples.

A short rules document — the scale, the permitted variants, what never nests inside what, which component owns page padding — is a small file that changes output quality substantially.

What the work becomes

The assembly compresses; the decisions do not. What belongs in the system, what a new pattern should be called, whether this generated layout is actually right — those stay human, and they matter more than before, because each one now propagates through everything produced against it.

Frequently asked questions

Does this mean designers are replaced?

It means the repetitive assembly work compresses. Deciding what the system contains, what the rules are and whether a result is right remains a design job — and becomes more leveraged, because those decisions now propagate through everything generated.

What format should the system be in?

Anything structured and current. A tokens file, a component manifest with names and props, a short rules document. Beautiful documentation that lives only in a design tool is not readable by the thing doing the assembly.

Sign in to react.
Share this post.

Comments

Sign in to join the conversation.

No comments yet. Be the first.

Don't miss this

You might also like

System prompts: the instructions the user never sees
System prompts: the instructions the user never sees
29/08/2026 — admin@byqreal.test

A system prompt sets the rules for everything that follows. Written well it is the cheapest quality improvement availabl...

Structured output beats parsing prose
Structured output beats parsing prose
27/08/2026 — admin@byqreal.test

If your code needs a value from a model, ask for JSON and validate it. Regexing an answer out of a paragraph is a bug wa...

An agent is a loop, not a personality
An agent is a loop, not a personality
23/08/2026 — admin@byqreal.test

Strip away the branding and an agent is a small loop: call the model, run the tool it asked for, feed the result back, r...