Atlassian ADS and Rovo evidence profile
Type: System profile
Status: Active
Canonical for: Factual interpretation of preserved Atlassian ADS and Rovo evidence
Depends on: corpus/atlassian/
Two linked changes
Atlassian explicitly links product-facing AI patterns (Rovo UI/Dialect) to an agent-facing “context engine” built above ADS. This is the corpus’s clearest organizational statement that a design system must serve humans and AI as consumers while also containing AI product patterns.
Four-question relevance
- Intent: documented as design intent, constraints and flexibility.
- Structure: documented as machine-readable schema files, structured content, skills and decision context; the key schema is not publicly inspectable here.
- Retrieval: documented through ADS MCP, skills, templates and portable Markdown.
- Constraint/evaluation: reported evaluations show accuracy/efficiency gains, but tasks, scoring and raw results are unavailable.
AI product design
- Documented: Rovo Dialect is part of Atlassian Design Language and defines how AI-powered moments look and behave through principles, foundations, components and patterns. It aims to be identifiable as AI while remaining integrated with Atlassian. Rovo UI overview
- Documented: The preserved Rovo surface includes AI interaction guidelines, button, skills tag, generative border, color, elevation, icon, illustration, motion, and voice/tone. Rovo documentation directory
- Documented: Rovo AI patterns distinguish Rovo branding and general AI-powered features. Rovo patterns
- Implemented: The public frontend mirror contains Atlassian design-system packages, tokens, components, types, stories and tests. It provides direct implementation evidence for ADS generally and some AI/Rovo surfaces, though the corpus does not show a public one-to-one implementation for every Rovo guidance page. Frontend mirror
- Documented: Rovo agents are configurable teammates grounded in Atlassian and connected knowledge. Agent product documentation
Public Rovo guidance is strongest on identity, expression and AI interaction. Detailed reusable patterns for consequential approval, undo, audit history and long-running autonomous actions are not comparably visible.
Context-engine architecture
- Documented: Atlassian defines an AI-native DS through four claims: AI can understand it; AI can build with it; AI patterns are part of it; and AI helps maintain system health. Context-engine article
- Documented: The expanded stack includes structured content files, ADS MCP, unified code-generation templates, ADS skills and portable Markdown. It is intended to connect Rovo, Figma Make, Replit and the design-system website.
- Documented: Atlassian reports evaluation results: 52% accuracy improvement in AI calls, 34% faster ADS-specific tasks, 26% fewer AI tooling calls and 16% fewer AI tokens. The preserved public article does not include the evaluation protocol or raw data, so these are organizational claims, not independently reproducible results.
- Documented: Accessibility is intended to be encoded into structured context and skills, not left solely as human-readable guidance.
- Inferred: The public repository snapshot makes conventional APIs and tokens inspectable; the most novel context artifacts and ADS MCP are described more completely than they are inspectable in this corpus. That is a documentation/implementation evidence gap.
Composition
- Documented: Atlassian explicitly moves from isolated components toward compositions, layouts and intent that AI can understand.
- Documented: ADS describes a three-tier composition model: core DS, cross-product platform patterns, and app-specific components. Innovation may move upward from app to platform/core.
- Documented: Panel embodies intentional constraints: a responsive page shell, an outer component built from primitives/tokens, contextual subheaders, and purposeful slots that allow content freedom without losing rhythm or accessibility.
- Documented: Atlassian later reframes the hierarchy as a constellation/web of decisions and decision trees. This admits that composition is contextual rather than a simple parent-child taxonomy.
- Inferred: Slots and tiering encode some composition structure, while the described context layer aims to encode selection intent. The public evidence does not yet demonstrate a stable, general machine-readable composition schema.
Debt, governance and accessibility
- Documented: Atlassian uses its labelling-system redesign as explicit evidence of accumulated design/technical debt: hacks, customizations and siloed solutions had diverged across apps. The replacement adds hierarchy, native interactivity and semantic accessible defaults.
- Documented: A date-time picker redesign corrected auto-opening on focus, reduced keyboard inputs, added semantic structure/labels and improved focus-ring/border contrast. This is conventional DS quality work that benefits agent use because better defaults become the generated defaults.
- Documented: Motion tokens are organized semantically so humans and AI can understand animation intent; tooling is still being developed.
- Implemented: The frontend mirror provides extensive tests, types, stories and tokens for ADS, but repository presence alone does not verify the context-engine evaluation claims.
Evidence assessment
Atlassian provides the strongest explicit integration of AI UX, semantic DS context, composition and agent tooling. Its conceptual and organizational evidence is unusually rich. Carbon and Storybook offer more directly inspectable AI-specific implementation; Atlassian’s novel context-engine claims therefore remain partly documented rather than fully implemented in public evidence.