Brand systems were built for one kind of reader: a person. A designer could pick up a brand book, absorb the spirit of it and apply judgement.

Brands now have a second reader, and it works in two directions. Outbound, AI tools are making the slides, pages and diagrams that go out under your name. Inbound, models are describing your company to customers, partners, candidates and investors, often before anyone visits your site.

Neither can follow a PDF. Hand an AI tool a brand book and it guesses, and the work comes back generic. Give a model an inconsistent picture of who you are and it gives back a vague answer, or describes you in a competitor's words.

Most brand systems were only built for the first reader.

We're building brand systems for humans and machines.

For humans:

All the usual deliverables are there: identity, brand book, Figma library, templates, example designs and asset files.

For machines:

The difference is in how it's built.

Brand guidelines are structured to be machine-readable: named variables, a clean component library, templates and canonical examples. Agents can read it directly.

Alongside it sits Go–MRK, our Machine Readable Kit: a set of files that live in whichever repo or tool the team works in:

• AGENTS.md: the entry point that tells any agent what the brand system is, what to read and the rules to follow

• DESIGN.md: the visual identity written as structured rules, covering colour roles, type hierarchy, layout, density and logo usage

• MESSAGING.md: what the brand says, written once: category, positioning, proof points with evidence, glossary, boilerplate and canonical answers

• constraints/: what not to show or say, each rule with its reason, including a claims register

• SKILL.md: the brand packaged as an installable Agent Skill, read by Claude, Codex, Cursor, VS Code and others

• tokens.json: every value (colour, type, spacing, radius) in the W3C Design Tokens format, exported from Figma variables and imported directly into code

• evals/: two test sets with scoring rubrics: outbound briefs that check what agents make, and inbound questions that check what assistants say about the company, re-run after any change

Every asset ships twice, once for each reader.

It works in two directions

Outbound: what AI makes for you. The kit governs what your team's AI tools generate on the brand's behalf. The slide, the page, the diagram comes back on-brand without a designer correcting it.

Inbound: what AI says about you. Models increasingly describe your company to customers, partners, candidates and investors, and they assemble that answer from how consistently you've described yourself. The kit fixes the category language, positioning and proof points once, and every site, deck and document the agents produce repeats them. Consistent input is the best lever there is on what models say back.

Most of the industry calls this GEO, and treats it as a publishing tactic: track the prompts, restructure the pages, chase the citations. That helps, but it works downstream. If the source is inconsistent, GEO just distributes the inconsistency more efficiently. The fix starts upstream, in the brand system itself.

The kit doesn't live on the public surfaces models read, but it supplies them: the website, owned channels and the third-party profiles that count most.

All of it is built on open standards, so it's portable, owned outright and not locked to any one tool.

Go–MRK is a new kind of brand system: designed for the people who choose you, and readable by the machines that now make for you and tell your story.