What Is a Design System? A Working Definition, With Two Real Ones

Artyom Sklyarov··7 min read

A design system is a maintained set of shared design foundations, components, patterns and guidance that helps a team build consistent interfaces. Its implementation can span design files, code and documentation. A style guide explains presentation; a UI kit supplies design assets; a component library supplies reusable implementation. A design system connects those resources with decisions about use, ownership and change. It reduces repeated decisions without removing the need for design judgment. Below: what goes in one, who needs one, what ours costs, and two real examples, a 115-icon language for an enterprise networking product and the system behind this site.

Four useful parts of a design system: tokens, reusable components, patterns and documentation, represented through design assets, code and guidance
Four useful parts of a design system: tokens, reusable components, patterns and documentation, represented through design assets, code and guidance

What is a design system, in simple terms

It is the answer to "how do we build screens here" written down once, in a form that machines and people both use. The colors are named, not picked. The type scale has steps, not sizes. The button has every state drawn and coded. The form has a rule for where the error goes. And a document says why, and what not to do.

The word "system" is doing the work. A collection of nice components is a kit. A system is when the components share the same decisions underneath, so that changing a decision, the accent color, the corner radius, the spacing unit, updates the components that consume that shared value.

Design system versus style guide versus component library

These get used interchangeably and they are not the same thing.

TermWhat it isWhere it livesWhat it cannot do
Style guideRules about look: colors, fonts, logo use, toneA PDF or a web pageBuild anything; it describes
UI kitDrawn components and screensA Figma or Sketch fileShip; it is a picture of the product
Component libraryReusable components, often with usage documentationA package or codebaseSupply every product-level decision by itself
Design systemShared foundations, components, patterns and governanceCode, design assets and documentationReplace research or decide which product to build

A useful warning sign is drift: the Figma button and the production button no longer match, and no one owns the discrepancy. A shared name alone does not keep them synchronized; the team needs a process for proposing, implementing and documenting changes.

What is in a design system: the four layers

Tokens. The raw decisions, named. This site has eight color tokens, five font families, five breakpoints, and a five-step radius scale (4, 8, 12, 16, and full), declared once in a theme block and referenced everywhere. Lime is the only accent and it is a token, so there is one place to change it.

Components. Buttons, inputs, cards, table rows, icons, each with every state drawn and coded: default, hover, focus, disabled, loading, error. The button on this site has four variants in a strict hierarchy, primary, outline, dark, and outline on dark, and nothing else is allowed to look like a button.

Patterns. Components in combination, with rules: how a form lays out, what an empty state says, how a page is structured. Every commercial page on this site is one pattern, a single offer template that renders eighteen different products from data. When the template improved, all eighteen pages improved in one deploy.

Documentation. The rules and the do-nots, written for someone who was not in the meeting. Ours is a file called DESIGN.md at the root of the repository, and the rule for both people and coding agents is the same: read it before any visual decision. It says which font carries the editorial accent, that dark heroes may use a colored shader and nothing else may, that there are no three-column icon grids and no uniform bubbly radius on everything. That last sentence is what documentation is for.

Two real design systems

An icon language for an enterprise product: 115 icons, one grid

The Apstra icon set: 115 icons across navigation, operations, intent-based analytics, charts, and AI, drawn on one grid with one stroke
The Apstra icon set: 115 icons across navigation, operations, intent-based analytics, charts, and AI, drawn on one grid with one stroke

Apstra is intent-based networking software from Juniper, now HPE: dense, expert, and full of concepts that have no ready-made pictogram. The icon system we design and steward for it is a design system in miniature: 115 icons across five product domains, navigation, operations, intent-based analytics, charts, and AI, drawn on one grid with one stroke weight, so that a glyph invented this month sits next to one from three years ago without a seam.

The systemic part is not the drawing, it is the dictionary. Every concept gets one icon, that icon gets one meaning, and new features draw from the dictionary before they add to it. When a team ships a screen with a new metaphor, the question is whether the metaphor already exists, and it usually does. That is the difference between an icon set and an icon system, and it is the same difference as between a UI kit and a design system. The 3D icons for the same product are the marketing layer on top of the same language.

The system this site runs on

The second example is the one you are reading. The tokens above, the four button variants, the offer template, and DESIGN.md supported our rebuild of fifteen commercial pages on a new template in a single day in early September 2026. This is a project observation, not a measured productivity study. Shared decisions reduced the number of choices in each page, and the agents doing a share of the assembly read the same rules file a designer would.

That adds another consumer of the system: a coding agent. Code components and written rules give it concrete references, but do not guarantee correct output. Design files can also provide context through compatible tools. In either workflow, inspect the result against the intended states and behavior.

Who needs a design system, and who does not

Consider investing when teams repeatedly solve the same component problem, interfaces drift or changes are hard to propagate. Team size and screen count are clues, not universal thresholds. Compare the maintenance cost with the repeated work the system would remove; start with the parts you actually reuse.

Keep the investment small when one designer-engineer is shipping a first version with few repeated patterns. Tokens, a component folder and short usage notes may be enough. Expand them when actual reuse or inconsistency justifies the maintenance cost.

What a design system costs

Public systems such as Google's Material Design, Shopify's Polaris, IBM's Carbon and GitHub's Primer offer useful references. Their scope should not become your launch checklist: begin with the components and patterns your product actually needs.

For a specific interchange reference, the Design Tokens Community Group format describes a shared token representation. It is a community specification, not a W3C Recommendation and not a substitute for component usage rules. The local counts and component examples in this article describe our September 2026 site, not a permanent inventory.

For a company with a product and no system, our published price is the Design System add-on, from $12,000 flat, two to three weeks: tokens, the core components with every state, the first patterns, and the documentation, delivered in code and in Figma, agreeing with each other. It attaches to a Design Sprint or runs inside a Design Partner retainer, where the system keeps growing as the product does. For context on where that sits in the wider market, the website design cost post has the tiers, and the earlier guide to design systems for startups covers how to keep one alive after it ships.

How to tell if you have one

Pick a screen the product does not have yet. Give the brief and system resources to a teammate who did not create them. Record which decisions they can reuse and where they need clarification. Questions about which blue, which button and where an error belongs reveal documentation or component gaps; questions about the user's task may need product research. This is a diagnostic exercise, not a pass/fail definition based on one person's speed.

Written guidance can also help when evaluating AI tools for UI/UX design. Test whether the tool can use your actual components and rules; a useful exploration tool may still require substantial implementation review.

For a concrete application, compare the SaaS palette contrast measurements with the typography decisions that affect usability.

A

Artyom Sklyarov

Founder & Designer at SUUR

More about Artyom →

Have a product idea?

Designed, built, and shipped in about 5 days.

Start a Project