If the person who designs a template already fixes almost every property by the act of making it, what does a brand contract still do? That question stopped me for a week — and cracking it forced me to admit something I hadn’t wanted to admit.
I’ve spent the last months trying to answer a stubborn parent question: can you make creativity computable? Not as philosophy — because generative AI is producing content at scale, and I want that content to be of the brand, not accidentally similar to it.
The answer, it turns out, is half. And precisely that half changes everything.
The setup: an explosion
At Foleon we’re building a new generation of layout — moving from a grid of columns to a genuinely two-dimensional canvas: rows, columns, spans, overlap. Creatively wonderful. Computationally an explosion. Where a template library used to fit neatly into ten folders, cataloguing simply breaks down once you can no longer enumerate the possibilities.
And when you can no longer enumerate something, you have to be able to describe it.
Layout is code, code is text
Every layout in our prototype is a small, readable text file: this tile spans column 1 to 12, covers two rows, contains a title. From that file, dumb reliable code computes a handful of properties — text-to-image ratio, whitespace, layering, regularity, density.
Every layout becomes a point in a multidimensional space. Not a metaphor; a calculation. And a language model can reason about those files the way it reasons about text — layout is as readable to an LLM as a sentence.
The brand is a box
Then: what does “fits our brand, for this purpose” actually mean?
The answer: a contract. If you’re informing, that pairs with roughly 70/30 text-to-image, medium whitespace, no overlap, a strict grid. If you’re inspiring, everything tilts — image carries, breathing room and layering are allowed.
A contract is a box in that same space — a bandwidth per property. And the creative question “does this layout fit here?” suddenly becomes a geometric one: is the point inside the box?
Where I got stuck
But when someone designs a template, they lock in most properties just by making the design. Proportions fixed. Structure fixed. Layering fixed. What is a contract still adjusting?
If the maker sets almost everything, contracts seem redundant. If contracts override the maker, you no longer have templates — you have raw material the machine deforms at will. Neither is what I wanted.
The way out came from recognizing that not all properties are the same kind of property.
Identity versus manifestation
Identity is what a design is: proportions, structure, layering. The maker freezes these. Contracts choose among them — they never deform what something is.
Manifestation is how a design shows itself in a context: mostly whitespace, mostly rhythm. Here the maker doesn’t specify a point but a range: “this layout may breathe between tight and airy.” The contract picks within that range. Two keys, both needed to turn the lock.
A template is therefore no longer a point in the space. It’s a point on the identity axes × a line segment on the manifestation axes. “Fits” means: the segment intersects the contract box. One design, manifesting properly differently under an informing contract than under an inspiring one — without ever needing two variants of the same template.
The half that failed — thankfully
Now the honest part: those handful of properties are necessary, but not sufficient. A Mondrian and a mess can produce identical measurements on every axis. The calculation guarantees on-brand and on-intent. It does not guarantee good.
That gap gets filled by something old-fashioned: human choice. Every layout a human approves goes into a library — and that library is, in effect, a knowledge base of taste. Human judgment, crystallized, the same way a knowledge base crystallizes text. The AI generates and tests; the human approves; the approved becomes knowledge; the knowledge shapes the next generation.
And that’s the conclusion I didn’t see coming: creativity does not become computable — the edges of creativity become computable. And precisely because of that, you can safely delegate the creative work within those edges to a machine. The machine can be boring or brilliant, but never off-brand: outside the lines does not exist for it.
Why this compounds
The implication stacks. Content teams that don’t make the identity/manifestation distinction will spend the next two years fighting their own AI. Teams that do get scale advantages that accumulate, because every human-approved output becomes fuel for the next generation.
The question I’d hand back to anyone working with generative content: which properties of your brand are identity, and which are manifestation?
Until you separate the two, you’re fighting your own AI. Once you separate them, it starts working for you.