Systems landscape

The ranking reflects public evidence, not organizational scale or product reputation.

Tier 1 — deep crawl

System	AI-related work	Formal DS?	Public components / code	Tokens / Figma	Agentic UX	AI-assisted design/dev	Evidence	Recommended depth
IBM Carbon	Carbon for AI, AI presence, label/explainability, AI variants, chat, accessibility, MCP	Yes	Yes / yes; React, web components, AI Chat	Public AI tokens; public Figma kits referenced	Limited; strongest on generated content, chat and control/revert	Yes; Carbon MCP	Very high, Level 1 across multiple surfaces	Full crawl
Microsoft Fluent + Microsoft agent guidance	Copilot UI Kits, content engineering, AI-first interaction architecture, agent lifecycle/accessibility guidance	Mostly yes; agent guidance spans Fluent and Microsoft Learn	Public Fluent code; AI kit implementation visibility varies	Fluent tokens and design resources; Copilot kits listed	Strongest public formal guidance found	Indirectly; less public DS-specific MCP evidence than IBM/Atlassian	High Levels 1–2; some assets gated	Full crawl with boundary mapping
Atlassian Design System	Rovo Dialect/patterns plus ADS “context engine,” structured files, ADS MCP, skills and generation templates	Yes	Core ADS public; public AI code surface less clear	Core tokens/resources public; Rovo asset access varies	Product agents exist; public UI patterns focus on Rovo identity/moments	Strong formal 2026 program	High Level 2, with implementation claims needing repository verification	Full crawl
SAP Fiori / Compositional Design System	Long-running AI/Joule guidance plus machine-readable composition logic used by Joule Work generative UI	Fiori: yes; compositional layer extends it	Fiori controls public; compositional implementation not publicly inspectable	Fiori resources available; compositional metadata/tokens unclear	Yes conceptually: cross-system agents and dynamic workspaces	Yes for generative composition; not a public coding-agent MCP	High guidance; production claim, limited code	Full crawl, clearly separate public vs claimed internals
Figma + Storybook ecosystem	MCP, Code Connect, component manifests, agent rules, story tests	Infrastructure serving many DSs, not one DS	Yes / yes	Figma variables and Code Connect; Storybook manifests	Not primarily	Direct and central	Very high Level 1 implementation evidence	Deep thematic crawl as enabling infrastructure

Tier 1 source maps

IBM Carbon

IBM Carbon
├── Carbon for AI guidelines
│   ├── introduction and AI presence styling
│   ├── AI label and explainability popover
│   ├── AI variants of core components
│   ├── AI chat
│   ├── accessibility
│   └── Figma and developer resources
├── Component documentation
│   ├── AI label (usage/style/code/accessibility)
│   ├── table/form/input/modal/tag/tile AI presence
│   └── revert behavior
├── Code
│   ├── carbon monorepo
│   ├── carbon-ai-chat
│   ├── Storybooks
│   └── tests/accessibility status
├── Foundations
│   ├── AI color tokens in themes
│   └── iconography/style elements
├── Carbon MCP
│   ├── overview and setup
│   ├── tools/resources and prompt behavior
│   ├── carbon-mcp repository
│   └── generated-code evaluation/examples
└── Design resources
    └── Figma v11 all-themes AI-presence modes

Microsoft Fluent and agent guidance

Microsoft
├── Fluent 2
│   ├── foundations, tokens and core component libraries
│   ├── Copilot UI Kits (web, iOS, Android)
│   └── content engineering
├── Microsoft Design
│   ├── AI-first system direction
│   └── Copilot interaction architecture / Throw & Catch
├── Microsoft 365 Agents design guidance
│   ├── human-centered lifecycle
│   ├── accessibility and inclusion
│   ├── Adaptive Cards for agent responses
│   └── responsible agent design
├── HAX Toolkit
│   ├── 18 human-AI guidelines
│   ├── patterns and examples
│   └── workbook
└── Public implementation check
    ├── Fluent React v9 / web components
    ├── token packages
    └── determine which Copilot assets map to shipped code

Atlassian Design System

Atlassian
├── Rovo UI / Rovo Dialect
│   ├── principles
│   ├── foundations (color, motion, illustration, icons)
│   ├── components
│   └── AI/Rovo patterns
├── ADS context engine
│   ├── structured context files
│   ├── ADS MCP server
│   ├── agent skills and portable Markdown
│   ├── generation templates
│   └── semantics / design intent
├── ADS core
│   ├── components
│   ├── tokens
│   ├── accessibility
│   └── composition patterns / Panel
├── Product surface
│   ├── Rovo agents
│   └── agent lifecycle/permissions evidence
└── Repositories
    └── locate public packages, context artifacts and MCP implementation

SAP Fiori and Compositional Design System

SAP
├── AI and Joule Design
│   ├── generative AI principles
│   ├── trust, safety and sustainability
│   ├── prompt guidance
│   ├── explainable AI
│   ├── feedback
│   └── matching, ranking and recommendations
├── Platform variants
│   ├── web
│   ├── iOS
│   └── Android / mobile resources
├── Compositional Design System
│   ├── relationship logic
│   ├── structured semantic metadata
│   ├── guardrails / valid combinations
│   ├── experience models
│   └── learning signals
├── Joule Work
│   ├── Spaces
│   ├── generative UI
│   └── cross-system orchestration
└── Evidence boundary
    ├── public Fiori controls and packages
    └── identify what compositional schemas/code remain private

Figma and Storybook enabling stack

Design-system agent infrastructure
├── Figma MCP
│   ├── design context extraction
│   ├── variables and code syntax
│   ├── Code Connect snippets / prop mappings
│   ├── custom instructions
│   └── code-to-canvas / canvas-to-code boundary
├── Code Connect repository
│   ├── React / HTML
│   ├── SwiftUI
│   └── Jetpack Compose
├── Storybook MCP
│   ├── component manifest
│   ├── documentation retrieval
│   ├── story instructions
│   ├── story tests
│   └── AGENTS.md guidance
└── Documentation delivery
    ├── zeroheight MCP
    └── compare direct docs, manifest and MCP surfaces

Tier 2 — useful supporting sources

System / source	Why useful	Limitation
Singapore Government Design System	Rare public 2026 example combining AI components, tokens, Figma MCP, Code Connect and skills while explicitly noting human review and current MCP deployment limits	Smaller surface and some direction is prospective
WordPress / Gutenberg	Shipped DS-specific MCP package exposing Storybook-derived component docs, examples and tokens	Focuses on coding-agent consumption, not AI product UX
Google PAIR	Foundational, research-backed AI product design methods and patterns	Not a formal Material Design AI extension and predates agentic wave
zeroheight	2026 survey provides field-level adoption data; shipped MCP demonstrates docs distribution	Vendor incentives; not itself a product design system
OpenAI agent system/safety material	Production evidence for confirmation, supervision and constrained action	Not published as a formal design system
ServiceNow practitioner research/writing	Concrete agentic UX patterns around consent, audit, undo and escalation	Located material is practitioner guidance, not public DS policy

Tier 3 — interesting but weak evidence

• AI2, Better Design, DesignSystems.one and independent DS MCP servers: useful experiments and terminology signals, but adoption, quality and independent validation are weak.
• Salesforce Lightning, Adobe Spectrum, GitHub Primer, GitLab Pajamas, Shopify Polaris and Oracle Redwood: important systems with AI-heavy products, but discovery did not find a coherent public AI-related DS surface comparable to Tier 1.
• UAE Design System: has an AI documentation assistant and AI-related stories, but the assistant is a way to query the system rather than evidence that the system itself supports AI experiences or agents deeply.
• Basis “Compass and AI”: a current public in-progress pattern area mentioning uncertainty, latency and override, but not yet enough implementation evidence.

Exclude from deep crawl

• Systems whose only evidence is an “AI-powered,” “AI-ready” or “agent-native” claim without inspectable guidance, code, tokens, component APIs, research or a credible implementation case.
• Famous systems should remain excluded until a direct AI/design-system link is found; product AI documentation alone is insufficient.