Vincent

Strategy & systems

One brand, two readers

A brand guideline was written for people who already understood the brand. AI is the first reader that knows nothing it isn't told.

Client
Foleon
Period
2021 — 2026
Type
Brand management for humans and AI
Role
Brand Kit, Brand governance, Product vision
PEOPLE HOLDmeaning · reasons · examplesMACHINES HOLDvalues · ranges · checksIDENTITYwho we arewhy we exista few hard no’sKNOWLEDGEwhat we can provewhat matters, and whyfacts with sourcesVOICEhow we soundsounds like usranges on dialsLOOKhow we appeartastebasics, styles, a matrixBEHAVIOURhow content movesfeelpatterns and timingsCONTEXTfor whom, for whatwhat this moment needsa region per purposeJUDGEMENTwhen to cross the linewhen outside is bettera log of every noEvery recorded “no” moves the line a little to the right. It never reaches the edge.

A brand book works because its readers fill the gaps. A designer who reads “confident, but warm” knows what to do: they have absorbed the brand from colleagues, from the work on the wall, from years of being corrected. An AI model has none of that. It reads exactly what is written, and fills every gap with the average of everything it has seen.

So the question for brand management changes. Not “is our brand described?”, but “which reader can use which part?”

Seven layers, two readers

A brand is not one thing. It is a stack, from what hardly ever changes to what changes per page:

  • Identity — who we are, and what we will not do.
  • Knowledge — what we know, and what we can prove.
  • Voice — how we sound.
  • Look — how we appear: colour, type, space, image.
  • Behaviour — how content moves: navigation, transitions, motion.
  • Context — for whom, and for what purpose, right now.
  • Judgement — when to deliberately do something else.

Every layer can be written for two readers. For people: meaning, reasons, examples. For machines: values, ranges, checks. Brand management for humans and AI is deciding, layer by layer, where the line between the two runs.

Where the line runs

In the middle, machines hold almost everything. Look and behaviour are the most computable parts of a brand. A colour is a value, a type scale is a set of steps, a transition is a duration and a curve. Written as values, they cannot be gotten wrong without breaking a rule you can see.

Knowledge, voice and context are shared. A machine can hold facts with their sources, and a voice as positions on a few dials with a band around each. It can hold context as a different region of the same space per purpose: informing is not inspiring, but it is still the same brand (Same space, different region). A person still hears whether it sounds like us, and knows which fact matters to this reader.

At the edges, the line moves to people. Identity is the reason behind every other rule; written down in full, it turns into a slogan. A machine can hold the few hard no’s, not the why. Judgement is the decision to cross a line on purpose, and that cannot be a rule: if you could state it, it would already be inside the boundary instead of the thing that judges it (Who holds what).

The look layer, worked out

This is where my own product work sits. As creative director I built Foleon’s Brand Kit as the platform’s reusable foundation. Years later, with agents on the horizon, it became the test case for one question: what does an agent need to read before it can make anything on brand? Three rules made the look machine-readable.

Basics have no meaning. A colour, a size, a spacing step: values without semantics. “Primary text colour” is not a basic — it is a rule that points at one. Colour modes reference basics; they never copy them.

Look in the kit, meaning in the document. The kit owns every text style, plus a matrix of which semantic tag each style may carry: one default, a few allowed. The author picks a style, the tag follows. The right look on the wrong tag is invisible to a reader and fatal to a machine.

Look and behaviour, apart. They change for different reasons, at different rates, by different hands — so they are governed separately and meet in the document.

THE KIT — THE LOOKbasics: values without meaningAaAaAatext styles: the rulesDISPLAYTITLESECTIONEYEBROWBODYTHE MATRIX — THE DECISIONwhich style may carry which tagH1H2H3PDisplayTitleSectionEyebrowBodydefault tagallowedTHE DOC — THE MEANINGH1H1PH2Ppick a style —the tag followsAn agent is exactly as brand-safe as the rules it can read.
Look in the kit, meaning in the document: the kit owns the styles, a matrix decides which style may carry which meaning.

What it asks of a brand team

  • Write every layer twice. Once as meaning for people, once as rules for machines. Where the second is impossible, say so. That is the human part, not a gap.
  • Make deviation visible, not impossible. Local exceptions stay allowed. They are shown, counted and, where it matters, locked.
  • Nothing publishes itself. A proposal — from a person, an import or an agent — becomes a rule only when someone decides.
  • Close the loop. Every “no, not that” that gets recorded moves a piece of the line from people to machines. It never reaches the edge, and it shouldn’t.

That is the brand team’s new job: deciding which rules an agent can read, and protecting the part that should never become one.

Related writing: The edges of creativity are the part that becomes computable and Every organization has a knowledge fingerprint.

All strategy work