AI SYSTEMS & DESIGN ARCHITECTURE

Mosaic: The AI Enterprise Design System

Translating static design system foundations into machine-readable contexts for AI design interfaces—empowering Product Managers to generate design system compliant mockups directly via natural language prompts.

Open Design AI Interface - PM Prompt-Driven System Mockups
Self-Service PM Mockup Interface (Prompt-to-UI) Claude Context Ingestion
Primitive Color Scales and Token Architecture
10-Stop Primitive Color Scales & Token System Global Variable Architecture
Enterprise Layout Blueprints and Master-Detail Archetypes
Enterprise Layout Blueprints & Master-Detail Archetypes High-Density Schemas
Role & Scope
Design System Contributor & AI Design Engineer
Platform
JAGGAER Enterprise SaaS Ecosystem
Tools & Stack
Figma, Open Design / Claude, JSON Tokens

Executive Summary

Building and maintaining UI consistency across complex enterprise B2B procurement and supply chain workflows requires rigid component governance and flexible authoring tools. As a Lead Product Designer dedicated to JAGGAER Direct, I focused on establishing scalable design token architecture, defining system governance, and bridging visual foundations into AI-assisted design environments.

By translating our design system standards into structured prompt contexts for AI design tools, we established a self-service pipeline where Product Managers could input natural language requirements to generate design-system-compliant mockups instantly, streamlining rapid prototyping and significantly reducing handoff overhead.

The Challenge: Scaling System Governance Across Teams

Enterprise platforms like JAGGAER manage vast, data-heavy interfaces—multi-step sourcing wizards, granular data tables, dynamic approval matrices, and complex dashboard configurations. While visual libraries establish baseline consistency, traditional handoff models faced structural friction as we transitioned from static, canvas-bound Figma components to an AI-ready, Claude-driven design system context.

Four Platform Stacks, One Central Design System
SINGLE SOURCE OF TRUTH
Mosaic
The JAGGAER Design System

Shared tokens, component Blueprints, accessibility rules, & machine-readable AI context schemas.

Powering UI Across All Products
JAGGAER Direct
Sourcing & Direct Spend
JAGGAER Indirect
Procurement & eProcurement
JAGGAER Advantage
Supplier & Network Portal
JAGGAER AI
Autonomous & Conversational
Design System Evolution Pipeline
Legacy Canvas
Static Figma Kits

Manual component detachments, visual inconsistencies, static canvas spec handoffs.

Transformation
Structured Token Engine

Standardized primitives, semantic JSON schemas, and structural layout blueprints.

AI Architecture
Claude AI Context Layer

Machine-readable rules empowering prompt-based, system-compliant UI generation.

1. Token & Context Disconnect

Translating visual Figma variables into code-level semantic tokens required manual verification and dev translation without machine-readable context.

2. Prototyping Bottlenecks

Constructing realistic, data-dense enterprise prototypes manually was time-consuming for designers and inaccessible to PMs wanting to sketch initial feature flows.

3. System Drift

Early product feature wireframes created during rapid discovery frequently used non-standard components, leading to redesign cycles before AI guardrails.

System Architecture & Blueprint Governance

Standardizing primitive token scales, component blueprints, and full-page layout archetypes established the programmatic rules required for both human designers and AI generators.

JAGGAER Direct Primitive Scales & Token Architecture
Tier-2 Semantic Mappings, Accessibility Specs, and AI Agent Tokens
Figure 1.0 & 1.1: 10-Stop Primitive Color Scales, Semantic Mappings, & AI Runtime Tokens v4.5 Architecture Spec
  • Token Hierarchy: Restructured variables into a three-tier model:
    Global Primitives: 10-stop scales across blue, gray, teal, red, green, and amber color families.
    Semantic Tokens: Contextual bindings (e.g., --color-action-primary, --color-status-conform).
    AI Behavioral Tokens: Runtime state directives (e.g., --ai-state-thinking, --ai-state-confidence-high).
Buttons System Interactive States & Anatomy
Buttons Accessibility Standards & Production Code
Enterprise Layout Systems & Page Archetypes
Figure 2.0: Component Anatomies & Standard Layout Archetypes (Master-Detail, High-Density Grid, AI Workspace)
PM WORKFLOW INNOVATION

Prompt-Driven PM Mockup Generation

One of the primary goals of this architecture was bridging the gap between product management requirements and design system compliance. By indexing design system tokens, layout blueprints, and component schemas into machine-readable prompts, we empowered Product Managers to generate fully compliant UI concepts directly from natural language.

Self-Service PM Mockup Interface driven by Claude context prompts
Figure 3.0: Prompt-Driven UI Generation Interface & Structural Token Rules Open Design / Claude Integration
Input & Context

Structured System Prompting

Product Managers input natural language feature stories (e.g., "Create an audit grid with status flags and primary action buttons"). The context layer automatically restricts output to approved tokens and component archetypes.

Speed & Compliance

Zero-Drift Self-Service

PMs get immediate visual validation during discovery without creating off-brand wireframes or waiting for dedicated design sprint allocation for preliminary concepts.

MEASURABLE VALUE & SCALING

Impact & Strategic Outcomes

Transitioning to machine-readable token schemas and AI-assisted design contexts fundamentally reshaped cross-functional product delivery across all four platform stacks.

85%
Faster Concept Discovery

PM discovery mockups generated in minutes rather than days of manual canvas work.

100%
Token Compliance

Zero non-standard color or spacing overrides in AI-generated interface prototypes.

3x
Engineering Handoff Speed

Direct mapping to frontend code JSON schemas eliminated manual UI translation backlogs.

Metric / Dimension Traditional UI Handoff AI System Integrated Workflow
Prototype Velocity Hours/Days in visual canvas tools Minutes via direct PM prompt entry
Token & Scale Adherence Manual review across design/dev review meetings Enforced at prompt context boundary
Cross-Platform Sync Disparate design libraries across 4 product teams Single source of truth via unified Mosaic tokens
Dev Alignment Static specs translated manually Generated layouts map to production tokens
Cross-Functional Empowerment

Democratized Prototyping

Product Managers now construct high-fidelity, system-compliant feature flows independently during early discovery. This freed product designers to focus on high-impact architecture, complex interaction workflows, and system governance rather than production mockups.

Ecosystem Governance

Eradication of System Drift

By constraining AI prompt outputs through structured JSON token rules and layout blueprints, the platform eliminated rogue styling and off-brand UI components across all four disparate product tech stacks.