Skip to content

Design system: the complete guide, 2026 edition

Definition, atomic design, design tokens in the DTCG format, Figma and React components, governance, budget, and design systems that AI agents can read. This guide is for product designers, product owners and front-end leads preparing a design system, and for digital directors looking into one before a tender.

Published by VOID. Updated on .

Key takeaways

  • A design system runs from the Figma file to the code: tokens, tested components, usage rules, documentation and governance.
  • Since October 2025, design tokens have a stable format (DTCG 2025.10), published by a W3C community group but not a W3C Standard.
  • Accessibility is handled in the component (WCAG 2.2 AA), but conformance is proven page by page.
  • In 2026, a design system must also be readable by coding assistants, through the Figma and Storybook MCP servers.
  • Without a pilot product and an owner for the run, a design system drifts within months.

Definition

What is a design system?

A design system is the shared set of principles, design tokens, interface components, usage rules and governance that lets several teams design and build consistent digital products, from the Figma file to production code.

In practice, a design system contains six things:

  • Principles: the choices that guide decisions (clarity over density, accessible by default…).
  • Tokens: design decisions stored as variables (colours, typography, spacing).
  • Components: buttons, inputs, cards, dialogs, designed in Figma and built in code.
  • Patterns: combinations of components for common needs (form, data table, sign-in).
  • Documentation: when to use each element, and when not to.
  • Governance: who decides, who contributes, how versions are released.

These elements stack up, from the most abstract to the most visible:

  1. Tokens

    Design decisions (colours, typography, spacing), written once as variables.

  2. Components

    Buttons, fields, cards: built with the tokens, in Figma and in code.

  3. Products

    Website, customer area, app: assembled from the same components.

  4. Agents

    Coding assistants read tokens, components and rules to generate on-brand screens.

From token to agent: each level builds on the one before. A token change flows into components, then into every product, and AI agents read the same rules as your teams.

Design system, style guide, UI kit, component library: the differences

Differences between a brand style guide, a UI kit, a component library and a design system
CriterionBrand style guideUI kitComponent libraryDesign system
ContentsLogo, colours, typefaces, toneComponent mock-upsCoded componentsPrinciples, tokens, Figma and coded components, patterns, documentation
MediumPDF, brand siteFigmaCode (React, Twig…)Figma, code, documentation site
Question answeredWhat does the brand look like?What does the interface look like?How is the interface coded?How do all teams build the same interface?
OwnerCommunications teamDesign teamFront-end teamDedicated team plus contributors, with a process

Stakes

Why build a design system, and when not to

The expected benefits are well known. They only show up once the system is adopted:

  • Consistency: the same experience on the website, the customer area and the app.
  • Speed: teams assemble existing components instead of redrawing and recoding them.
  • Maintenance: a colour or component change happens in one place.
  • Onboarding: new designers and developers find the rules written down.
  • Accessibility: criteria are handled once, in each component.

We do not put figures on these gains: they depend on your starting point, and the percentages that circulate never say what scope they were measured on. Measure your own (see Measuring success).

A full design system is not always the answer. For a single product and a small team, a component library and a few tokens are enough. The full system makes sense when several teams, products or brands share the same building blocks, or when accessibility and consistency become obligations.

Method

Atomic design: structuring your components

Atomic design was proposed by Brad Frost in a post dated 10 June 2013. The idea: we are not designing pages, we are designing systems of components. The method describes five levels.

  1. Atoms

    Basic elements that cannot be broken down further: button, input, label, icon.
  2. Molecules

    A few atoms working together: a search field with its button.
  3. Organisms

    Complete interface blocks: a header with logo, menu and search.
  4. Templates

    The structure of a page, with its organisms, without real content.
  5. Pages

    The template filled with real content, to test the system in context.

The method has limits. The line between molecule and organism causes long debates, and the terms mean little to product teams. Many teams keep the idea (compose from small to large) and simplify the names: foundations, components, patterns, layouts. That is the structure used in the rest of this guide.

Foundations

Design tokens

A design token is a design decision stored as a named variable: a colour, a font size, a spacing value, a radius, a shadow, an animation duration. Tokens are the single source that both Figma and the code read.

Three tiers of tokens

  • Primitive: the raw palette (color.red.600, spacing.4). Never used directly in components.
  • Semantic: the role (color.action.primary, color.text.error). This is the tier designers and developers work with.
  • Component: settings specific to one component (button.primary.background), when it needs to differ from the general role.

Naming, modes and multiple brands

Name by purpose, not by value: color.text.error survives a change of red, red-500 does not. Modes (light, dark, high contrast) and brands are handled by changing the values of semantic tokens, without touching components. That is how a group can roll out one system to several subsidiaries.

The Design Tokens Community Group format (2025.10)

The Design Tokens Community Group (DTCG), a community group hosted by the W3C, released the first stable version of its format, numbered 2025.10, on 28 October 2025. The text says so plainly: it is not a W3C Standard and it is not on the W3C Standards Track. It defines a neutral JSON format ($value, $type and $description properties), references between tokens, theming and several colour spaces. According to the group, more than ten tools and open-source projects support or implement it, including Style Dictionary, Tokens Studio, Figma and Penpot. The format text recommends the .tokens and .tokens.json file extensions.

Here is an extract in the 2025.10 format. It replaces the example from the first edition of this guide:

{
  "color": {
    "$type": "color",
    "red": {
      "600": { "$value": { "colorSpace": "srgb", "components": [0.827, 0.137, 0.165], "hex": "#d3232a" } },
      "800": { "$value": { "colorSpace": "srgb", "components": [0.561, 0.078, 0.094], "hex": "#8f1418" } }
    },
    "action": {
      "primary": { "$value": "{color.red.600}" },
      "primary-text": {
        "$value": "{color.red.800}",
        "$description": "Red text on light backgrounds: contrast of at least 7:1"
      }
    }
  },
  "spacing": {
    "$type": "dimension",
    "md": { "$value": { "value": 16, "unit": "px" } }
  }
}

primary and primary-text hold no value of their own: their $value is an alias that points to a primitive token, using the {color.red.600} syntax. The type is declared once on the group and inherited by its tokens.

The production pipeline

Figma variables (or Tokens Studio) export tokens as JSON. A transformation tool such as Style Dictionary turns them into CSS variables, a Tailwind theme, and iOS and Android resources. The JSON file is versioned with the code: a token change goes through review, just like a component change.

Components

Components, patterns and templates

Each component in the system is defined by:

  • Variants: primary, secondary, outline, subtle.
  • States: default, hover, focus, active, disabled, loading, error.
  • Sizes: a small number, aligned with the spacing scale.
  • Props: enough to cover real cases, not so many that anything can be changed.
  • Accessibility: role and ARIA attributes, focus management, keyboard navigation, target size.

Patterns combine components for a specific need. Three deserve particular care in banking, insurance and service products:

  • Forms: labels always visible, error messages tied to their field, information already entered in the same process filled in automatically or offered for selection, with some exceptions (WCAG 2.2 criterion 3.3.7).
  • Data tables: sorting, filters, pagination and mobile reading, without losing headers for screen readers.
  • Authentication: no mandatory cognitive test, copy and paste allowed for codes and passwords (WCAG 2.2 criterion 3.3.8).

Templates set the structure of standard pages: home, list, detail, sign-up flow. They are mostly used to test the system before rolling it out.

Documentation

Structuring and documenting a design system

A structure that works in most cases:

  • Foundations: principles, tokens, colour, typography, grid, spacing, iconography, motion.
  • Components: one page per component.
  • Patterns: forms, tables, navigation, authentication, empty and error states.
  • Content: tone, microcopy, date and amount formats, rules per language.

Every component page follows the same outline: purpose, when not to use it, anatomy, variants and states, props, illustrated do and don't rules, accessibility criteria, code example and a link to the Figma component. A component without documentation goes unused, or gets misused.

Write for agents too. Coding assistants read the documentation: explicit rules ("use Button variant primary for a single main action per screen") work better than a page of screenshots.

Tooling

Tools in 2026

Design system tools in 2026, their role and notes
ToolRoleWorth knowing
FigmaComponents, variants, variables, Dev ModeCode Connect maps each component to its code; the MCP server exposes variables, components and layout to agents
Tokens StudioToken management in Figma, sync with a Git repositoryListed by the Design Tokens Community Group among tools supporting the 2025.10 format
Style DictionaryTurns tokens into CSS, Tailwind, iOS and Android resourcesListed by the Design Tokens Community Group among tools supporting the 2025.10 format
StorybookDocumentation, interaction, visual and accessibility testsMCP server in preview: component docs and test runs for agents
Tailwind CSSUtility classes fed by the tokensThe theme is generated from the tokens file
Accessible headless librariesAccessible behaviour without styling (menus, dialogs, tabs)Saves recoding focus and keyboard handling
Drupal (Single Directory Components)Twig, YAML, CSS and JS for a component in one folderStable in core since Drupal 10.3; fed by the same tokens as React

Sources: Figma MCP server documentation, Storybook MCP server documentation and Single Directory Components guide on Drupal.org. Choose tools after the audit, and favour the ones your teams already use if they can read tokens.

AI

Design systems and AI

Coding assistants now generate part of the screens. Without machine-readable rules, they produce plausible interfaces that sit outside the system: approximate colours, recreated components, invented spacing. The design system becomes the reference they check.

What the agent needs to read

  • Tokens, in the DTCG format and with semantic names.
  • The list of components, their props and their usage rules, written as sentences.
  • A rules file at the root of the repository: which components to use, what is forbidden, how to test.

The connectors

The Figma MCP server passes the variables, components and layout of a design to the coding assistant. Code Connect adds the actual import and usage of the component in your code. The Storybook MCP server, in preview, gives access to component documentation and lets the agent run tests, including accessibility checks.

Components for AI interfaces

A product that embeds an assistant needs new components: generation states (in progress, partial, error), fallback answers, handover to a human, confidence cues, source citations. And the notice telling users they are talking to an AI: Article 50 of the EU AI Act has required it since 2 August 2026, unless it is obvious, according to the European Commission. We cover the design of these journeys in our Harnessed UX service.

AI in governance

AI can also watch over the system: spot detached components in Figma, hard-coded colours, gaps between design and code. Its proposals go through the same reviews and tests as human ones. That is the principle of harness engineering.

Accessibility

Building accessibility into components

A component made accessible once is accessible everywhere it is used. It is the most concrete argument for a design system. The technical target is level AA of WCAG 2.2, whose new criteria directly affect components: focus not obscured, minimum target size, alternatives to dragging, accessible authentication, redundant entry.

The regulatory frame depends on the market:

A design system helps, but it does not make a site conformant on its own: conformance is proven page by page, through an audit. For the method, read our practical WCAG guide.

Organisation

Governance: who decides, who contributes

The core team

A team, even a small one, owns the system: it maintains it, reviews proposals, trains the teams that use it and measures adoption. There are two models. Centralised: the core team builds everything. Federated: product teams contribute, the core team arbitrates and keeps things consistent. Most organisations move from the first to the second as the system grows.

The contribution process

  1. Proposal

    The need, the use case, a mock-up. Is there already a similar component?
  2. Core team review

    Accepted, merged with an existing component, or declined with an explanation.
  3. Design and documentation

    Figma component, variants, usage rules, accessibility criteria.
  4. Build and tests

    Coded component, Storybook story, interaction, visual and accessibility tests.
  5. Code review

    By the core team and a member of the team that made the proposal.
  6. Release and announcement

    New version, release notes, message to the teams.

Versioning and deprecation

  • Major (2.0.0): a breaking change.
  • Minor (1.5.0): a new component or option, no breaking change.
  • Patch (1.4.3): a bug fix or minor improvement.

A component never disappears overnight: it is first flagged as deprecated, with its replacement and a removal date. A regular ritual (proposal review, demo of what is new) keeps the link with the teams.

Roadmap

Launching a design system: the six-step roadmap

  1. Audit

    Interface inventory, Figma files, front-end code, team interviews. You know what exists and what diverges.
  2. Scoping

    Scope of the first version, components prioritised by actual use, core team appointed.
  3. Foundations

    Tokens (primitive and semantic), typography, grid, iconography.
  4. First components

    The components that cover most of your screens, in Figma and in code.
  5. Pilot product

    A first product adopts the system. You measure and fix before scaling.
  6. Scale and run

    Other products follow. The system moves into ongoing maintenance, with a budget and an owner.

Do not skip the last step: a design system delivered without an owner for the run decays within months.

Measurement

Measuring success

Three families of indicators, to track from the pilot product onwards:

  • Adoption: share of products using the system, share of the interface built with its components, rate of detached components in Figma.
  • Progress: documented components, contribution proposals processed, time between a request and its release.
  • Performance: design and build time for a standard screen, interface and accessibility defects found in testing, satisfaction of the teams using it.

Measure before you start. Without a baseline, no gain can be demonstrated.

Budget

Budget and timelines: what drives the cost

The cost of a design system depends on five factors:

  • The number of products and brands to cover.
  • The state of what exists: clean Figma files or not, components already coded or not.
  • Platforms: web only, or web, iOS and Android.
  • The level of documentation expected.
  • The run: who maintains the system after delivery, and with what commitment.

For a quote on your case, see our design system service.

Examples

Public design system examples

  • Système de Design de l'État (DSFR): the French government's design system. Circular 6411/SG of 7 July 2023 (in French) makes it mandatory for every new website or mobile app of central government departments, prefectures, embassies and devolved State services; State operators may use it with the approval of the Government Information Service (SIG). It deliberately limits customisation in favour of a shared experience. A reference for anyone working with French public bodies.
  • Material Design 3 (Google): a model for tokens and dynamic theming, documented for the web, Android and Flutter.
  • Carbon (IBM): an open-source system built for enterprise software products, with React and web components.
  • Polaris (Shopify): since 1 October 2025, its unified components ship as web components for the admin, point of sale and checkout. Proof that a design system can change technology while keeping its rules.
  • Atlassian Design System: detailed documentation on tokens, content rules and accessibility.

Pitfalls

Mistakes to avoid

  • Starting too big: begin with the most used components, then iterate.
  • Building without a pilot product: a system never tested on a real product will not be adopted.
  • Skimping on documentation: an undocumented component goes unused.
  • Being too rigid: plan a route for edge cases, or teams will work around the system.
  • Forgetting governance: without contribution rules, the system turns back into a pile of variants.
  • Leaving accessibility for later: build WCAG 2.2 in from the first components.
  • Forgetting the run: without an owner and a maintenance budget, the system decays within months.

FAQ

Frequently asked questions

What is a design system, in one sentence?

A design system is the shared set of principles, design tokens, interface components, usage rules and governance that lets several teams design and build consistent digital products, from the Figma file to production code.

What is the difference between a design system and a brand style guide?

A brand style guide sets the visual identity: logo, colours, typefaces, tone. A design system turns that identity into reusable, tested building blocks: tokens, coded components, usage rules, documentation. The style guide says what the brand looks like; the design system says how to build an interface with it.

What is atomic design?

A method proposed by Brad Frost in 2013 to structure an interface in five levels: atoms (button, input), molecules (search field), organisms (header), templates (page layouts) and pages (templates filled with real content). It helps teams think in systems of components rather than pages.

What is a design token?

A design token is a design decision stored as a named variable: a colour, a font size, a spacing value, a radius. Tokens are written once, in a neutral format such as the Design Tokens Community Group (DTCG) format, then transformed for Figma, the web, iOS and Android.

How should design system documentation be structured?

In four parts: foundations (principles, tokens, typography, grid, iconography), components (purpose, anatomy, props, states, accessibility, code examples), patterns (forms, tables, authentication) and content (tone, microcopy). Every page says when to use the element and when not to.

How long does it take to launch a design system?

A useful first version, limited to tokens and the most used components on a pilot product, takes weeks; a system covering several products takes months and is never really finished. The schedule depends on the number of products, what already exists and the platforms.

Do you need a design system for a single application?

Not always. For one product and a small team, a component library and a few tokens are often enough. A full design system, with governance and documentation, pays off when several teams, products or brands need to share the same building blocks.

Can an AI use a design system?

Yes, if it is machine-readable: tokens in the DTCG format, explicit names, documentation that explains when to use each component. The Figma and Storybook MCP servers give coding assistants direct access to components, variables and documentation. AI proposals should then go through the same reviews and tests as human code.

About this guide

This guide is published by VOID, a digital agency operating since 2005 with production teams in Casablanca and Agadir. It was fully rewritten in September 2026 to cover DTCG 2025.10 tokens, MCP servers, WCAG 2.2 and the European Accessibility Act. Meet the team on our agency page.

Planning a design system? Let's look at what you have

Audit, tokens, Figma and React library, governance and run: explore our service, or tell us about your products.