Shape Grammar System. A visual system for my writing. I drew the grammar; AI turned it into a system that maps every post.

The problem

My writing had no images. Twelve posts, twelve title cards in a pastel box.

Stock illustration felt wrong for posts about judgment and process. Generated images would have looked like everyone else's. What I wanted was a system, not pictures: something with rules, where a post's image says something about its argument, and where post thirteen fits without anyone redrawing the first twelve.

Three shapes. Five ways to relate.

I started with the grammar and one reference graph. Every glyph is built from a circle, a triangle and a square, joined by one of five relations. Relations compose, so any glyph can become part of a larger one. That's what makes it a system and not a set of icons.

Primitives

Circle●The signal. A topic, an entity, a point of attention.

Triangle▲Direction and change. Something moving or rising.

Square■Structure. A framework, a rule, a base to build on.

Relations

ConnectA — BPeers side by side, joined centre to centre with an ink line and two nodes.

StackA / BA rests on B. Reads as built on, or depends on.

NestA ⊂ BA sits inside B, drawn in ink. Reads as contains, or is part of.

OverlapA × BA outlined across B. Reads as shared ground between two things.

AnchorA | BA small A tethered above a larger B. Reads as derived from, or flags.

Notation

A — B   connect
A / B   stack
A ⊂ B   nest
A × B   overlap
A | B   anchor
( )     group
{ }     one to many

Fill carries the shape, ink carries the relation.

Fill carries the shape, ink carries the relation. Lines, nodes and nested marks are always ink; bodies are always fill. That one rule keeps every glyph readable, as a small cover or a full hero.

Nine pairs, five relations, 45 glyphs.

Each row is an ordered pair A · B. Order matters: a triangle on a square is not a square on a triangle.

Seeing every combination at once is how you test a system. If one cell reads wrong, the rule behind it is wrong, and it's wrong everywhere.

Every ordered pair of shapes against every relation
A · B	ConnectA — B	StackA / B	NestA ⊂ B	OverlapA × B	AnchorA \| B
● · ●Circle · Circle
● · ▲Circle · Triangle
● · ■Circle · Square
▲ · ●Triangle · Circle
▲ · ▲Triangle · Triangle
▲ · ■Triangle · Square
■ · ●Square · Circle
■ · ▲Square · Triangle
■ · ■Square · Square

Relations chain into larger glyphs.

A finished glyph behaves like a single shape, so it can be connected, stacked, nested or anchored again. Brackets group a sub-glyph.

These compounds are the reference patterns. Every post glyph is composed the same way.

Chain▲ — ● — ■A sequence. Cause, signal, response.

Pillar(▲ / ■) — ●A built thing linked to the signal it answers.

Totem● / ▲ / ■Three layers stacked. A hierarchy read top-down.

Core link(● ⊂ ■) — ▲A contained signal pointing to a change.

Branch● — {▲, ■}One source, two outcomes.

Hub■ — {●, ●, ▲}A structure that gathers many signals.

Shelter(● ⊂ ▲) / (■ ■)A shared roof over two bases.

Eclipse▲ | (● × ■)A flag raised over shared ground.

Beacon▲ / (▲ ⊂ ●)A change rising from a change held inside a signal.

Lantern● / (▲ ⊂ ■)A signal carried on a structure that holds a change.

Keystone■ / (● ⊂ ▲)A structure set on a change that contains a signal.

Relay(▲ / (▲ ⊂ ●)) — ●A Beacon passing its signal on to the next.

Where AI did the work

The grammar and the reference were mine. Everything after that I built with AI, and that's where the scale came from.

Mapping meaning to form. Turning a post into a glyph is a translation problem: read the argument, find its core claim, name the parts, decide how they relate. AI reads every post with the same rules applied every time and proposes a formula for each, with a one-line reading of why it fits.

1 · The post says“By getting things wrong faster and more often.”

2 · Mapped to● — ▲ — ■ — ●Idea, action and structure joined in a closed loop.

3 · Drawn as

Building out the system. From the reference, AI generated the matrix and the compounds above: every pair against every relation, then the chains that combine them. That's the exhaustive part a system needs and a person rarely finishes.

Scaling it. The glyphs live in code, not image files. Colour comes from the same tokens as the rest of the site: each post takes the pastel of its topic. A check runs on every build. Every post has a glyph, no glyph breaks the grammar, no two posts share a formula. When I publish a new post, the writing skill proposes two formulas as part of the process. I pick one.

What stays mine

AI proposes; I decide.

What the shapes mean, that a triangle is change and that nesting means "part of", is a design decision, and it's the thing that makes the glyphs mean anything. So is saying no to a mapping that's technically correct but reads wrong. A formula can follow every rule and still miss the point of the post.

Every post so far

The gaps we never wrote down■ — ○ — ■

Before you hand your design system to an agent(■ ⊂ ■) — ▲

The Human Creativity Benchmark● | (■ ■ ■)

The Learning Loop● — ▲ — ■ — ●

AI for better outcomes▲ — ▲ — ▲

AI in Design 2026 Report(● ● ●) ⊂ ■

It's Not the AI○ × ■

Three critics for every flow{▲, ▲, ▲} | (■ — ■)

Write to think first, then prompt■ — ● — ▲

Discovery Over Delivery● / ■

Your skills.md● | (▲ ⊂ ■)

Runnable Intent● — (● ⊂ ▲)

The rules

Three primitives, five relations. Nothing else.
Five shapes at most.
No text inside a glyph.
One formula per post, never repeated.
Colour comes from the post's topic, never picked per glyph.