DEVOURED
Design Systems that Document AI
Only 26 out of 156 public design systems meaningfully document AI, though leaders like IBM, AWS, and Microsoft independently converged on four core principles for AI-powered experiences.
What: A survey of 156 public design systems revealed only 26 address AI. Leading teams, including IBM, AWS, GitLab, and Microsoft, independently converged on four core principles: always mark AI-generated content, explain it in layers, keep humans in control, and design for failure. IBM Carbon for AI and AWS Cloudscape are noted for their robust guidelines.
Why it matters: This analysis reveals that while the industry is rapidly adopting AI, design system practices are lagging, yet leading companies are independently establishing crucial guidelines for ethical and usable AI. This convergence signals emerging best practices for integrating AI responsibly into user experiences.
Takeaway: If you manage or contribute to a design system, begin incorporating guidelines for marking AI-generated content, explaining AI decisions, ensuring human control and override capabilities, and explicitly planning for AI failures and recovery paths.
Deep dive
- A survey of 156 public design systems found only 26 address AI, an increase from a year ago.* Leading companies such as IBM, AWS, GitLab, Red Hat, SAP, Workday, and Apple, despite no coordination, have converged on four core principles for documenting AI in design systems.* These four rules are: 1) Always mark AI-generated content, 2) Explain AI in layers (What, Why, How), 3) Keep humans in control, and 4) Design explicitly for failure.* The article introduces an "AI-readiness maturity model" for design systems, ranging from Level 0 (no AI guidance) to Level 5 (AI as system infrastructure, exemplified by Microsoft Fluent's Copilot UI Kits).* IBM Carbon for AI is praised for its tokenized AI identity, where an interactive AI label doubles as a trigger for layered explanations.* AWS Cloudscape offers detailed patterns for user-authorized actions, defining authorization levels, human-in-the-loop sequences, and rules against irreversible changes without explicit consent.* GitLab Pajamas emphasizes designing AI to be collaborative, not autonomous, with risk assessments and fallback rules for when AI confidence is low.* Red Hat PatternFly highlights the importance of a "value gate," questioning whether AI is truly needed before implementation, a principle also echoed by Apple and GitLab.* Critical gaps in most public design systems include guidance on AI confidence levels, source attribution, hallucination recovery, user correction loops, and multi-agent workflows.* The author stresses that the true value of a design system for AI lies not just in component creation, but in documenting when and how to use AI responsibly and what happens when it fails, treating AI documentation as living infrastructure.
Decoder
- Design system: A set of interconnected patterns and shared practices that provide consistency and efficiency in product development and design.* Explainability framework: A set of guidelines or tools designed to help users understand why an AI system made a particular decision or prediction.* Hallucination (AI): When an AI model generates information that is plausible-sounding but incorrect or fabricated, not based on its training data.* Copilot UI Kits: A specific set of UI components and patterns, developed by Microsoft Fluent, designed for integrating AI assistant experiences (like Copilot) into applications across various platforms.
Original article
Full article content is not available for inline reading.
Explore all 55 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi