AI-readiness audit

Is your design system machine-readable?

A scored diagnostic across six dimensions, a live test where an agent builds a known screen from your system unaided, and a prioritized fix list.

Who it's for

01 Product teams whose AI coding tools — Claude Code, Cursor, Copilot, Figma MCP — ship off-brand, inconsistent UI, even with a "complete" design system.
02 Agencies inheriting a client's system and deciding whether to build on it or rebuild it.
03 Anyone about to hand a design system to an automated pipeline and who wants a number before, not an incident after.

What the audit measures

Each dimension is scored 0–100 with evidence, impact and fix. The AI-readiness score is the weighted average.

01 Token architecture Semantic layering, naming, modes, primitive vs. alias separation. gray-700 names a value; text-secondary names a job — and the job is what an agent has to guess at.
02 Component parity Figma components ↔ code components, 1:1. Code Connect coverage. Components that map, not resemble.
03 Naming consistency Drift across design, code and docs. One name per thing, one convention.
04 Machine-readability Is there a source of truth an agent can read? CLAUDE.md, token export, structured docs — not PDFs, not Slack threads.
05 Agent consumability The live test: an agent is asked to build a known screen from your system, unaided. What it gets right, what it guesses.
06 Governance How changes propagate and who owns the truth. Without this, every fix drifts back.
diagnostic bands
<40 little an agent can read
40–70 partial — agents guess on the gaps
>70 most of what an agent needs is present

What you get

01 Diagnostic report — score per dimension, evidence for each finding, and a remediation list split into quick wins and structural fixes.
02 AI-readiness score — one 0–100 number, benchmarked, so progress is measurable after the fixes.
03 Live test record — the screen the agent was asked to build, what it produced, and where it broke.
04 Findings session — a walkthrough with your team: the findings and the recommended path.

How it runs

01 Kickoff — read access to Figma and the component repo; scope confirmed.
02 Audit — the six dimensions, plus the live consumability test.
03 Score and prioritize — findings ranked by leverage, not by count.
04 Delivery — report plus findings session.

You provide read access to the Figma file(s) and the code repo, and your team's time for kickoff and findings.

What happens after

Most systems don't need a rebuild. They need the token layer re-architected, the core components mapped, and documentation an agent can execute. That is the Brand → system engagement. Teams with a steady flow of brands or products use Standing capacity: a recurring slot in the pipeline.

Or the audit says you're fine. That happens too, and it's a good outcome.

offer / agencies Brand → system, for agencies writing Writing a CLAUDE.md your design system respects
Design systems your AI can read.
Led by Diego Nogueira ↗