DS in AI: Problem exploration and recommendations

Objective

Investigate the intersection of AI tools and design system usage to understand current gaps, validate user needs, identify potential solutions, and provide actionable recommendations for making Pajamas more accessible to AI systems.

Open questions

Problem Definition

What does it mean to make the AI aware of our design system?

  • Clarify the original context and intent behind this suggestion
  • Identify specific pain points or opportunities that prompted this idea
  • Determine which AI tools are in scope (GitLab Duo, ChatGPT, Claude, others)
  • Understand which use cases matter most:
    • AI-assisted code generation and suggestions
    • Duo providing better design system feedback in MR reviews
    • Prototyping by designers, PMs, and engineers using AI
    • Other scenarios

What problems do users experience when AI tools lack design system awareness?

  • Examples of when AI-generated code would have benefited from being Pajamas-aware
  • Assess whether this is a significant problem worth solving

Solution Exploration

Goal 1: Can Duo be aware of our design system for smarter MR reviews and code suggestions?

  • Research how AI code assistants can be trained or configured to understand specific design systems
  • Explore GitLab Duo's current capabilities and extension points
  • Identify what information Duo would need access to and in what format

Goal 2: Can we make our design system more "AI-friendly"?

  • Determine what makes a design system consumable by AI tools (documentation updates, additional code examples, the addition of machine-readable docs, endpoints or files, or other any other changes)
  • Assess whether waiting for AI tools to naturally learn from updated documentation is sufficient, or if proactive steps are needed

Goal 3: Can prototypers explicitly instruct AI to use the GitLab design system?

  • Explore how users can provide context to AI tools (prompts, configuration, custom instructions)
  • Investigate MCP (Model Context Protocol) servers or similar mechanisms for providing design system context
  • Determine if this overlaps with Goal 2 or requires separate solutions

Competitive and Technical Research

What are others doing?

  • Research how other design systems handle AI tool integration
  • Identify industry best practices and emerging patterns
  • Review technical approaches (structured data formats, API integrations / mcp servers, training data, encouraging AI-indexing, etc.)

Deliverables

  1. Problem Statement: Articulation of the problem, its importance, and impact on users
  2. Scope Definition: Identification of which AI tools, use cases, and user personas are in scope and why
  3. Solution Analysis: Overview of potential approaches with pros, cons, and applicability to each goal
  4. Recommendations: List of solutions with rationale for which approach(es) provide the most value
  5. Feasibility Assessment:
    • High-level technical feasibility evaluation for recommended solutions
    • High-level effort estimate
  6. Implementation Approach: Rough scope breakdown

Out of Scope

  • Actual implementation of any solutions
  • Training or fine-tuning AI models
  • Creation of prototypes or proofs-of-concept
Edited by Paul Wright