AI for friction, humans for meaning. Raw output is never finished work. Verify before trusting. Disclose where it matters. Skepticism welcome when specific.
Standing recap.
The setup
Forty seconds with a model gets you
A positioning line. A palette. Three taglines.
All competent. All interchangeable with every brand in the category.
"And competent-but-interchangeable is the definition of brand failure."
The thesis
The economics of this week
Clients never bought assets. They bought judgment.
AI just made assets nearly free — which makes judgment the entire remaining product.
Career framing, not tool framing.
Today
Today's promise
A mini brand system for something you actually care about — with every decision that's yours marked and countable.
"✏️ marks on the sheet. We count them at share-out."
Deliverable
You will leave with
A mini brand sheet: positioning (de-souped) · voice + banned words · type/colour rationale · asset & export checklist
Show the prepared Toy Cathedral sheet for 20 seconds.
Today's map
Today's map
Questions, positioning, systems — 40 min
Stress tests + the contrast trap — 15 min
Demo: Toy Cathedral — 35 min
Lab: your micro-brand — 70 min
Share: de-souped lines + ✏️ counts — 40 min
Breaks protected.
Part 1
Questions before assets
~0:10.
The difference
Amateur vs. professional
Amateur
"Design me a logo."
Professional
Interrogates first: audience, occasion, budget, what it must never feel like, the secret competitor.
"The model is good at ASKING these when told to. It cannot answer them — the answers live in your head and your client's."
In real work
"They asked things nobody else asked"
wins pitches.
The question list is the first thing a client sees of your process.
True at every studio the students will work at.
Part 1, continued
Positioning
The sentence everything hangs on
~0:18.
The formula
For [narrow audience], [name] is the [category] that [genuine difference].
Everything downstream argues for this sentence — or doesn't belong.
Type, palette, voice, layout: all evidence for the sentence.
The problem
AI positioning is soup
"For music lovers, Toy Cathedral is the band that brings energy."
Competent. Fluent. Describes ten thousand bands.
Read it flat. Let it die in the air.
The move of the week
The de-souper
Strike every word a competitor could also claim.
What survives is your difference. If nothing survives — the brand doesn't have one yet. That discovery is billable.
"Not a prompt problem. A strategy discovery." Runs live in the demo.
Part 1, continued
Systems, not assets
~0:26.
A brand is rules
The system's four parts
Voice: 4 adjectives, each with "this not that"
Type: 2 pairings with rationale tied to positioning
Palette: colours + where each is allowed
Layout logic: grid habits, spacing personality
"Swatches are decoration. Usage rules are the system."
Quality control
The rationale test
Reject any type rationale that could justify anything.
✗ "Clean and modern" — justifies everything ✓ "The slab's bluntness matches 'playful, not precious'" — a decision
Demand rationale cite the positioning. Recurs in the demo.
Copy
The tone spec
Your 4 adjectives + this-not-thats + a banned-words list:
elevate · seamless · vibrant · discover · [yours]
"Without a tone spec you get LinkedIn voice. A good banned-words list is a brand asset."
Break
Break — 10 minutes
Stage WebAIM + Google Fonts tabs.
Part 2
Stress tests + the contrast trap
~1:00.
Stress tests
A system that only works on the A3 poster isn't a system
Make AI attack it: phone size · sticker size · support-slot poster (your name at 10%) · the cynical Dublin gig-goer
Then revise the single weakest element.
"A stress test that ends in a redesign was a tantrum, not a test."
Set your watches
The contrast trap
Models will assure you a colour pair passes accessibility.
Models estimate. WebAIM calculates. You'll watch the model lose this argument in about 30 minutes.
Setting up the demo's reliable laugh + permanent habit.
Production
Where hours come back
Asset checklists with specs · export QA lists · naming conventions · alt text drafts · style-guide assembly
Pure friction. Fully delegable. Specs always verified.
"Platform dimensions drift constantly — never trust a model's recall of them."
Demo
Toy Cathedral
Weak idea → campaign kit
~1:15. Prompts from DEMOS file. Tabs ready.
Demo · move 1
Interrogation first
I want "a logo and poster for my synth band, Toy Cathedral." Before you make anything: ask me the 6 questions a senior brand designer would ask first. Number them. Don't answer them yourself.
I answer live. My answers are the strategy.
Answer with personality: venue size, audience, sound references, budget (broke).
Demo · move 2
System, not assets
Watch the de-souping
Positioning ×5 → de-soup one live → voice → type rationale (reject one) → palette with usage rules → [GUESS] marks visible
De-soup on screen with the room striking words aloud. Reject one rationale: "could justify any font — regenerate."
Demo · move 3
Stress test
Phone. Sticker. Support slot. Cynic.
Weakest element → one targeted revision.
Room picks the weakest element with you.
Demo · move 4
The contrast moment
The model said the pair passes.
WebAIM, live. Let's see.
The reliable beat: estimate vs real ratio. If it passes today: "Today it passed. The habit is the checker, not the trust."
Demo
Tie-off
What just happened
Questions → strategy (mine) → system drafts (its) → opinions and rejections (mine) → maths (the checker's).
Every ✏️ on the kit is a decision. Count yours today.
Show the finished prepared kit page 20 seconds.
Break
Break — 10 minutes
Lab next: your micro-brand.
Standing brief staged: "Notions," the late-night toastie hatch.
Lab · 70 minutes
Brand System Stress Test
Interrogate + position (20) → System (20) → Stress test (15) → Checklist + sheet (15)
Standing brief: a late-night toastie hatch beside a Dublin venue, called "Notions"
Parts + times on board.
Lab · phase 1
20 min
Interrogate + position
6 questions answered in writing → 5 positioning lines → de-soup → keep the survivor.