The Infrastructure

This is what machine-readable culture looks like

CLAUDE.md is not documentation. It is the operating system for a six-agent team. Every page on this site was generated by agents governed by this single file.

Context > Probability

Why one file changes everything

Without CLAUDE.md, AI generates probable output — generic, off-brand, unfocused. Whatever the model thinks a “design systems architect” sounds like. With CLAUDE.md, AI generates contextual output — on-brand, strategically aligned, voice-correct.

This is Jesse Gardner's “Context > Probability” thesis from Into Design Systems 2026, made tangible. One file. Six agents. Every page on this site.

The File

CLAUDE.md, annotated

Key sections from the actual file, with annotations explaining why each section exists and which agents consume it.

Brand & Positioning
## Brand & Positioning

### Title
**Design Systems Architect**
— subtitle: *NorthStar Prototyping | Agentic Design*

### Key Differentiators
- **NorthStar Prototyping** — Exploratory AI prototyping
  to shape vision early when possibilities are widest
- **Agentic Design Systems** — Machine-readable,
  code-based design systems that AI agents can consume
- **One-person innovation engine** — AI makes one person
  as productive as a discovery team

This section tells every agent who Lincoln is and what makes him different. The Content Strategist uses it to position every piece of writing. The Stakeholder Simulator uses it to test whether a pitch lands. Without this, agents generate generic output. With it, they generate Lincoln.

Content StrategistStakeholder SimulatorUX Researcher
The INC Framework
### INC Framework (Ideate < Narrate > Create)
The core methodology:
- **I (Ideate)** — Discovery-first NorthStar prototyping
- **<N> (Narrate)** — The agentic design system
  infrastructure that amplifies both directions.
  N = domain knowledge encoded as agentic context.
- **C (Create)** — Production code, content, shipping

N is the expansion layer — not a step between I and C,
but the amplifier that radiates into both:
  I <N> C

This is the intellectual foundation. Every thoughts page ties back to this framework. The Content Strategist structures arguments around it. The UX Designer uses it to justify navigation decisions. It's not documentation — it's the operating model.

Content StrategistUX DesignerUX Researcher
Writing Voice
## Writing Voice
- Direct, confident, no corporate fluff
- British English spelling (organisation, prioritise)
- Uses "I" not "we" — this is a personal brand
- OK to mention Claude Code by name
- "Vibe coding" used casually, never as a title
- Bold claims backed by career evidence, not hype

Six rules that control the voice across every page. Without these, the Content Strategist writes like a generic marketing bot. With them, it writes like Lincoln. The FED Developer ignores this section entirely — and that's correct. Boundaries work both ways.

Content Strategist
Key Projects
## Key Projects (for reference in content)
| Project   | Role                    | Company          |
|-----------|-------------------------|------------------|
| Fun Lab   | Design Associate Dir    | Dentsu/Merkle    |
| Helix     | Chief Design Officer    | Telstra Health   |
| PenCS     | Design Director         | PenCS            |
| Breville  | Design Systems Lead     | Breville         |
| Adelaide  | Design Systems Lead     | Merkle           |
| Red Roost | Design Systems Spec     | NCGroup          |

Concrete proof points. When the Content Strategist makes a claim about multi-brand design systems, it can reference Fun Lab. When the Stakeholder Simulator plays a sceptical CTO, it knows Lincoln has Telstra Health credibility. Evidence, not assertion.

Content StrategistStakeholder SimulatorUX Researcher
Important Rules
## Important Rules for Editing This Site
- Keep the "proof-of-practice" framing — this site
  demonstrates agentic design, not just describes it
- The 50/30/20 split stays on website, off resume
- Don't use LINC acronym in hero — too obscure
- The design-system.html page is a killer differentiator
- All content should be generable from context files
- Footer states everything was generated via Claude Code

Guardrails that prevent agents from breaking the strategy. The FED Developer won't accidentally remove the proof-of-practice framing. The Content Strategist won't use LINC in a hero section. These rules encode decisions that were made once and shouldn't be revisited every conversation.

All agents
Perspective Shift

Same file. Six different readings.

Each agent reads the same CLAUDE.md but focuses on different sections. Same source of truth, different consumption patterns. This is the Clan culture argument made concrete.

CS

Content Strategist

Focuses on

Voice rules, brand positioning, INC framework, key projects for credibility

Ignores

Tech stack, CSS tokens, build configuration

FD

FED Developer

Focuses on

Tech stack, site structure, important rules about what not to break

Ignores

Writing voice, career arc, conference context

UR

UX Researcher

Focuses on

Target audiences, key differentiators, conference context, competitor positioning

Ignores

CSS properties, component structure, deployment config

UI

UI Designer

Focuses on

Design system tokens, colour palette, typography, spacing scale

Ignores

Writing voice, career narrative, project history

UX

UX Designer

Focuses on

Site structure, page hierarchy, INC framework for navigation logic

Ignores

Brand voice, visual tokens, deployment rules

SS

Stakeholder Simulator

Focuses on

Everything — needs full context to simulate realistic stakeholder reactions

Ignores

Nothing. Breadth of context is the point.

The Landscape

CLAUDE.md vs the alternatives

Other practitioners are solving the same problem differently. Nate Baldwin uses structured prompting with project-rules.mdc. Brad Frost advocates context files. Here's how CLAUDE.md compares.

CLAUDE.md
project-rules.mdc
Context Files
Scope
Brand + voice + architecture + governance
Technical rules + schema compliance
Design system + component specs
Persistence
Lives in repo, evolves with project
Per-session project rules
Separate context files per concern
Agent governance
Defines agent roles and boundaries
Constrains single agent behaviour
Informs but doesn't constrain
Brand voice
Fully encoded — tone, spelling, perspective
Not addressed
Not addressed
Cultural context
Career arc, methodology, positioning
Technical decisions only
Component usage only
Recursive proof
The file governs the agents that built the site that describes the file
Governs the tool that builds the tool
Documents the system
Meta

This page was generated by an agent governed by the file this page describes

The Content Strategist wrote this page's copy. The Content Strategist's agent definition references CLAUDE.md. CLAUDE.md defines the writing voice used on this page. The FED Developer implemented it. The FED Developer's agent definition references CLAUDE.md for tech stack rules.

The file is not just described here — it is actively governing what you are reading right now. If the tone feels consistent across every page, that's CLAUDE.md working. If the claims feel grounded in real projects, that's CLAUDE.md working. If the CTA feels like an invitation rather than a sales pitch, that's CLAUDE.md working.

When I write CLAUDE.md, I'm not writing documentation. I'm writing culture.

Want to build your own CLAUDE.md?

I help teams encode their culture, voice, and architecture into context files that AI agents can consume. The result: consistent output, governed creativity, and a team that scales without losing its identity.

/