How narrative expands both directions
Domain knowledge encoded as machine-readable context doesn't just sit in the middle. It actively amplifies what's possible on both sides.
N → I · Expanding Discovery
Narrative widens ideation
When the design system encodes domain knowledge, AI agents can explore possibilities that a human alone would miss. The narrative layer gives agents the context to ideate within real constraints.
- NorthStar prototypes grounded in domain truth
- AI explores the full possibility space, not just obvious paths
- CLAUDE.md as discovery context, not just build instructions
- Stakeholder language encoded—CEO, CIO, CTO each get their view
N → C · Accelerating Execution
Narrative accelerates creation
Machine-readable tokens, typed variants, and component contracts mean agents don’t guess—they compose with precision. The narrative layer eliminates ambiguity from the build process.
- Tokens as structured data, not design specs to interpret
- Component APIs that agents consume without hallucination
- Code Connect maps Figma to codebase—zero translation loss
- MCP endpoints expose the system as queryable infrastructure
I → N · Discovery Feeds the System
Ideation enriches the narrative
Every NorthStar prototype reveals domain patterns that feed back into the design system. Discovery doesn’t end at a handoff—it continuously enriches the machine-readable layer.
- Prototypes surface edge cases that update token contracts
- User research becomes encoded context, not slide decks
- New patterns graduate from exploration to system components
C → N · Production Validates the System
Code strengthens the narrative
Production implementation reveals which abstractions hold and which break. The build process is a feedback loop that makes the narrative layer more precise over time.
- Real usage data refines component variant APIs
- Performance constraints shape token decisions
- Agent-generated code validates machine-readability claims
What is an agentic narrative?
It's the encoding of domain knowledge, design decisions, and organisational context into formats that AI agents can consume and act upon.
Not documentation for humans. Not component APIs for developers. A structured narrative that gives machines the why behind the what—so they can make informed decisions rather than probabilistic guesses.
CLAUDE.md files, design tokens, typed component contracts, MCP server endpoints—these are all forms of agentic narrative. They turn a design system from a reference library into AI infrastructure.
// CLAUDE.md — agentic narrative in practice
Brand: Healthcare
Primary: #059669 // trust, clinical calm
Radius: 0.75rem // soft, approachable
Tone: professional, reassuring
// The agent doesn't just know the values.
// It knows WHY these values were chosen.
// It can make decisions that ALIGN
// with the domain intent.
Components:
Button: variant=primary|secondary|ghost
Card: elevation=sm|md|lg
Alert: severity=info|success|warning|error
// Typed. Constrained. Machine-readable.
// The narrative IS the infrastructure.One person. Full team capability.
When you hire me, you get an architect backed by a specialised agent team — each with defined expertise, boundaries, and context.
Lincoln Mitchell
DS Architect
UX Research
UX Design
UI Design
Visual
Motion
Content
FED Dev
Stakeholder
Lincoln Mitchell
DS Architect
+ 8 specialist agents
UX Research
UX Design
UI Design
Visual
Motion
Content
FED Dev
Stakeholder
The design system is the narrative.
The narrative is the infrastructure.
Let's talk about encoding your domain knowledge into systems that AI agents can actually use.