The Design Director Plugin

The Design Director Plugin

Twelve skills, six optional connectors, and one background agent. It captures your brand once, then makes everything you build look like it belongs to you instead of to whichever AI tool made it.

Download Plugin

the-design-director.plugin

⏱ 20 min read ● Beginner Version 1.0.0

Section 01

What is actually in the plugin

The Design Director is twelve skills, six optional connectors, and one background agent. This section is the inventory, so you can see the whole thing before you use any of it. Everything after it explains how the pieces work together.

You do not have to learn twelve skills. You say what you want in ordinary language and the right one picks itself up. The list below exists so you know what is available and can call something by name when you want to.

Getting your brand in

SkillWhat it doesSay this

design-setupCaptures your identity from a site, a logo, a deck or a brand kit, and says plainly whether you have a real brand, a conflicted one, or a template's defaults wearing your name"set up my brand"

brand-foundationBuilds one from scratch when there is not one yet. Palette solved for legibility, type pairing, shape language, spacing rhythm, offered as three genuinely different directions"I don't have a brand yet"

Deciding before you build

SkillWhat it doesSay this

design-directionTurns a vague request into decisions specific enough that the result could only be yours: the job, the reader's starting state, the proof that genuinely exists, the one action, and what gets left out"what should this page say"

Making the thing

SkillWhat it doesSay this

typographySets the words so a reader can see the structure without reading it. Computes the scale, leading, tracking and line length instead of eyeballing them"the text looks off"

colour-and-contrastPuts your colours where they belong, keeps text readable over photographs, handles dark mode, and catches anything carrying meaning by hue alone"check my colours"

layout-and-compositionGroups what belongs together, lines things up, counts the spacing arithmetic, and checks it survives a narrow screen"this feels cluttered"

imagery-and-visualsDecides what picture belongs, where it comes from, and whether it should exist at all. Most slot-filling images need deleting"what image goes here"

motion-and-interactionDecides what moves, how fast, and mostly what should not. Covers hover and tap feedback, loading, and the setting that lets people turn motion off"should this animate"

copy-and-microcopyWrites the buttons, errors, labels, empty states and small print nobody writes, checked against how you actually write"write the button labels"

Checking what you made

SkillWhat it doesSay this

anti-slop-passFinds what makes a finished piece read as machine-made, works out which of the four causes it is, and fixes it in order"this looks AI-generated"

design-critiqueA proper scored review across six dimensions, with a prioritised list and a note on what is already working and must survive the next revision"review this page"

redesign-and-refreshImproves something that already exists without breaking what people were relying on, and tells you when a redesign is not the answer"my website looks dated"

⚡

Eleven of the twelve carry a working check, not a checklist. Real arithmetic runs against your file: contrast ratios, colour vision simulation, spacing rhythm, type scale, page weight, voice match, and what disappeared between an old version and a new one. Section 10 covers what each one measures.

Section 02

The connectors it can use

Six connectors are declared with the plugin. Every one of them is optional, and every one only removes a step. Nothing in the pack requires a connector to produce its full output.

ConnectorUsed byWhat it saves you

Figmadesign-setupThe strongest capture route in the pack. Reads your variables and design-system rules, so colours, type and spacing arrive already named rather than inferred

Canvadesign-setup, imagery-and-visualsReads a stored brand kit, and generates the honest kinds of image: texture, pattern, illustration, diagram

Adobe for creativitydesign-setupReads a stored brand and its colour themes out of a creative-suite library

Gmaildesign-setup, copy-and-microcopyFinds something you actually wrote, so the pack can capture how you write without you going to look for a sample

Google Drivedesign-setup, imagery-and-visualsReaches your brand files, logos, decks and photographs where you already keep them

Notiondesign-setup, and anywhere a document is producedReads an existing brand or style document if one lives there, and can hold the brand sheet where your team will find it

Browsing is the seventh, and it is not in that list

Seven of the twelve skills get materially better when the session can look at a live page: design-critique, colour-and-contrast, layout-and-composition, typography, anti-slop-pass, motion-and-interaction and redesign-and-refresh. Several of their checks are simply impossible from source code. Real contrast over a photograph, what happens at phone width, keyboard order, whether motion stutters, and what a page said before somebody replaced it.

Browsing is a first-party browser extension rather than something you install from the connector list, so there is nothing to declare for it. Where it is available in your session, those seven skills pick it up on their own.

🔑

If you connect nothing at all, everything still works. Between them the twelve skills need only one of: a web address, a file, an image, or a conversation. Every artifact is produced as a file every time, and the pack never mentions a connector you do not have.

What it deliberately will not read a brand out of

Tools that generate work rather than store an identity are not brand capture routes, however design-adjacent they look. Reading a generated deck back in and calling the result a brand gives you a brand made of that generator's defaults, which is the exact failure this pack exists to prevent. Where such a tool is connected, the pack writes into it. It does not read a brand out of it.

Section 03

The agent that runs in the background

The plugin installs one agent, design-radar. It runs weekly by default, and you can ask for it any time by saying "what still needs doing on my designs".

It reads only your own brand file and the records the pack has written. It needs nothing connected, it opens nothing to form new opinions, and it never re-runs a skill's work. Its whole job is noticing patterns across records that no single run could see.

Blanks nobody filledThe pack leaves a marked blank rather than inventing a fact. That only works if somebody fills them in, and nothing else tracks whether they got filled. Ordered by how close each piece is to being published.

The same finding three timesA finding that turns up in three pieces is a brand or direction problem, not three separate faults. The radar names it, names the pieces, and names the one skill that fixes it at the source.

Undecided fields the work keeps needingUndecided is a legitimate state until the work needs that field twice. An undecided dark mode nobody has ever needed is fine and never gets raised.

Finished pieces never checkedAnything with a direction and no critique or slop pass recorded against it. Stated once per piece, without nagging, because you may have decided not to check it.

A quiet week is reported as a quiet week in one line. The radar does not manufacture work to look useful.

Section 04

Why everything you make looks the same

Ask any AI tool to build you a landing page and you will get something competent. The spacing will be reasonable. The colours will not clash. Nothing about it will be wrong.

And it will look exactly like the page it built for the person before you, and the one it will build for your competitor next week.

That is the actual problem, and it is worth being precise about it, because most people describe it as "it looks generic" and then try to fix it by adding things. Slop is not a list of forbidden elements. It is a structural property. The output has landed on the statistical centre of its category, where every individual element is defensible and the whole thing could belong to anyone.

Here is the test, and it takes five seconds. Take your name, your logo, and your colours off the thing you just made. Put a competitor's on. Would anything else have to change?

If the answer is no, you have the problem. Not because the work is bad, but because good taste is not a brand. Good taste is what every AI tool already produces, which is precisely why everything made this way looks like everything else made this way.

🎯

The core idea: The Design Director does not make your work more tasteful. It makes it more yours. Those are different goals, and only one of them makes you look different from the next business.

Everything listed in the three sections above exists to serve that one goal. The rest of this article is how you actually use it, starting with the only step that is not optional.

Section 05

Start here: set up your brand

Say "set up my brand" and nothing else. That is the entire instruction.

What happens next is not a form. The setup skill opens with one question, and it is deliberately not the question you would expect:

You: set up my brand Is there anything that already looks the way you want this to look? A website, a logo file, a deck, a brand kit, even a screenshot. Whatever's closest.

Not "what are your brand colours". Most people either do not know, know one of them, or name the colour they wish they used. What everybody can do is point at something.

Paste a link and the skill reads what your site actually paints: the typefaces resolved on your headings and your body text as two separate answers, the colours weighted by how much surface they cover rather than how often they appear in the file, the corner radius on your buttons and cards and inputs as three separate readings, and how you separate one thing from another.

The verdict you might not want

Before writing anything down, the skill decides what kind of brand you actually have. There are four answers and two of them are uncomfortable.

InheritA real identity, consistent across at least two places. It gets extracted exactly as it is, with no improvements.

SettleAn identity exists but your sources disagree. Your logo is one colour, your site another, your deck a third. Three versions of a brand is not a brand, it is three.

BorrowedWhat you have is a website builder's defaults with your logo on top. The fonts are the platform's, the palette is the theme's.

Nothing yetNo identity at all, which is fine and common. You still get your registers, your anti-references, and your voice captured.

Borrowed is the finding people least expect and most need. There is a specific test for it: take the logo off and put a different business's logo on. If nothing else would have to change, there is no brand underneath. That is usually the real reason everything you make looks like everything everyone else makes, and no amount of prompting fixes it.

⚠️

It will not quietly improve your brand. If your accent is a colour nobody would have chosen, it gets recorded as that colour. A brand you recognise and mildly dislike is worth more than a better one you have never seen, because only the first one actually gets used.

The question nobody asks you

Setup asks what you do not want to look like. Two or three things: a competitor, a category cliché, a look you are sick of. And then, for each one, the question that makes it usable: what is the specific thing that is wrong with it?

"Too corporate" constrains nothing. "Every one of them has a stock photo of people pointing at a laptop" removes a decision from every image brief the pack ever writes. These get checked as constraints on everything you make from then on, not as inspiration.

What you get at the end

A brand sheet. One page showing your colours and what each one is for, your type at the sizes it will really be used at, your shapes, and a worked example of the kind of work you actually make.

It is rendered rather than written for one reason: a wrong colour is invisible in a list of hex codes and obvious the moment it is painted. You are asked to look at it and say what is wrong, which takes ten seconds now and saves six weeks of things being built on a mistake.

Section 06

No brand yet? It builds one properly

If setup comes back Borrowed or Nothing yet, the next skill derives one. This is where the pack does something most tools do not.

Before generating anything, it asks what your brand can honestly be derived from. There are four answers, and the fourth is the one that matters:

SourceWhat it means

An anchorOne real fixed point exists. A logo colour, a material you work with, something printed you will not change. Best case.

The workNo anchor, but what you make imposes real limits. Food photography needs a ground that does not fight it. Dense numbers need room.

The negativeYou know what you refuse to look like. Derive by elimination. Weaker, and honest.

NothingNo constraint at all. The skill says so plainly rather than producing something confident and arbitrary.

That last row is not a failure state, it is the point. Anything generated with no constraint is a guess wearing the costume of a decision, and you would live with it for years. So the skill stops and asks three questions instead, any one of which converts the verdict.

The most useful of the three, and the one almost nobody is ever asked: where is somebody actually looking at this, and in what light? A phone outdoors and a monitor at night want different grounds, and the ground decides more about how the final thing looks than the accent colour does.

The palette is solved, not chosen

Here is the piece worth understanding, because it is why the output holds up.

Most tools pick a colour that looks nice, then check whether it is readable, then nudge it when it is not. The nudging is where a brand quietly becomes whatever passed the check.

This inverts it. The hue is the decision. The lightness is the solution. You choose the hue from your anchor, your work, or your elimination, because that is the part that carries meaning. Then the maths solves for the lightness at which each colour clears the contrast its job actually requires.

🔑

Why this matters to you: a palette built this way cannot fail its own accessibility checks, because it was constructed out of them. You never get handed a brand that breaks the first time someone tests it.

Three directions, not one

You get three genuinely different options, side by side, all showing the same worked example so you can actually compare them. There is a hard rule behind it: if two directions would differ only in accent colour, they are one direction.

Each one states what it costs you as well as what it does. The dark one is harder to print and harder to read outdoors. The airy one says less per screen. A direction with no stated downside reads as a recommendation, and this is not a recommendation. It is your choice and you live with it far longer than the conversation.

One of the three is deliberately the category standard played completely straight, built as well as the others. Partly because it lets you see how far the other two actually go, and partly because sometimes it is the right answer.

Section 07

Decide before you make

This is the step that is always skipped, and skipping it is why generated work is competent and interchangeable.

When you say "make me a landing page", a normal AI tool just makes one. It makes a tasteful one, which is the whole problem, because tasteful and generic is what you get when the only inputs are a category and a colour.

The direction skill turns that request into decisions specific enough that the result could only be for you. It asks at most four questions, and the interesting one is this:

You: build me a landing page for my new service How do people usually arrive at this? From a search, a recommendation, a card, an ad, a price comparison?

What state your reader arrives in

Not a demographic. The state of mind somebody is in the moment they land. Two people identical in every other way, one arriving from a recommendation and one from a price comparison, need completely different pages.

ColdDoes not know they have the problem. Needs to recognise themselves before you offer anything.

AwareKnows the problem, not the options. Needs the shape of the solution and roughly what it costs.

ComparingHas two other tabs open. Needs the difference, concretely, in the first sentence.

ReadyAlready decided. Needs the friction removed and nothing else.

There is a fifth, Worried: they want it, and something specific is stopping them. And a rule follows from all of it that runs against instinct: a reader further along needs less, not more. Giving somebody who already decided the full case for what they decided on reintroduces doubt they had already resolved. That is how a page talks a buyer out of it, and it is the most expensive mistake on the list because it loses the easiest customers.

What proof you actually have

Before any structure is decided, the skill inventories your evidence, in order of strength. Demonstration first, because showing the finished job beats everything. Then a specific outcome with a person attached. Then numbers you could defend. Then real third-party words.

This happens first for a reason: a piece with three real named results is built differently from one with none, and finding that out halfway through leaves holes in the layout that get filled with something invented.

💡

If you have no proof yet: the answer is not to make the claim louder. It is to make the ask smaller. A conversation, a quote, a look at the job. Somebody who will not buy from an unproven business will often talk to one, and the conversation is where the proof gets made.

Section 08

The six skills that make the thing

You will rarely call these by name. Ask for what you want and they pick themselves up. But it helps to know what each one is actually doing.

Typography

Sets your words so a reader can see the structure without reading it. The idea carrying it: hierarchy is made of five things, not one. Size, weight, tone, space, and position. A hierarchy built only from size is the tell, and it produces the page where three levels of big all compete and nothing leads.

It also computes what nobody can eyeball. Line height and letter spacing are properties of a size, not of a brand: the same typeface at 16px and 48px wants different values of both, and one value applied everywhere is the commonest reason type set by a system looks set by a system.

Colour and contrast

The palette was already decided, so this skill decides where each colour goes and how much of it there is. The discipline that carries it:

⚡

The accent's power is its rarity. One thing per screen should be the accent, and it should be the thing you want people to act on. A page where the accent is on the button, the links, the icons, the headings, and a decorative shape has no accent. It has a second body colour.

Layout and composition

Arranges things so what belongs together looks like it belongs together. One principle underneath all of it: space is a grouping signal, not a margin. Related things sit closer to each other than to anything else, and that gap difference does more structural work than any border or box.

Where everything is evenly spaced, nothing is grouped, and your reader has to do the grouping themselves. That is the most common structural failure in generated layouts, and no palette or typeface rescues it.

Imagery and visuals

Every image gets one question before anything is sourced: what is it for? Evidence, explanation, atmosphere, or nothing. That fourth answer is correct more often than the other three combined, and nobody ever reaches it.

This skill also asks the question that is worth more than every other piece of image advice: have you got photos of the actual work on your phone? The answer is nearly always yes, and most people do not think of them as usable because they are not professional. They do not need to be. A slightly imperfect photograph of the real thing beats a polished photograph of something else in every case, because it is the only image on your page a competitor could not also use.

Copy and microcopy

The buttons, errors, labels, and empty states nobody writes. These are the moments where somebody is stuck, confused, or about to act, and they are almost always written last by whoever is least interested.

The rule that fixes most of it: a button label is the outcome, not the mechanism. "Submit" describes what the software does. "Send the enquiry" describes what the person gets.

And the most common failure in small business writing, which this skill flips in one pass: sentences that start with "we" and "our". The reader has to translate those into their own situation before they mean anything, and most will not bother.

Motion and interaction

The only one of the six whose usual answer is no. Most things on a page should not move, and the question every proposed animation has to survive is whether it is telling the reader something. Feedback that a tap registered is information. A heading sliding in because headings slide in now is decoration with a delay attached.

Where motion is warranted, duration is the whole argument. Anything under about 100 milliseconds reads as an instant jump, and anything past roughly 400 reads as the page being slow. Small feedback belongs near the bottom of that band, and something entering the screen belongs near the top.

And one thing here is not a taste question at all. Large sliding or parallax movement genuinely makes some people nauseous or dizzy, and their operating system already carries a setting that says so. Every piece of motion the pack writes respects that setting, and anything that plays on its own can be stopped.

Section 09

Checking what you made

Two skills here, and they do deliberately different jobs.

The anti-slop pass finds what makes a finished piece read as machine-made, and fixes it. It runs the substitution test, then decides which of four problems you actually have, because the remedies are opposites and applying the wrong one makes things worse.

VerdictWhat it meansThe fix

FabricatedIt contains invented evidenceStrip it immediately, before anything else

GenericNothing is wrong, nothing is specificAdd specificity, never decoration

DecoratedVisual moves carrying no meaningRemove, then look again

BorrowedSomeone else's look in your coloursReturn to your brand

Notice that Generic and Decorated need opposite treatment. A generic piece feels like it is missing something, so the instinct is to add. That instinct is almost always wrong: what is missing is not volume, it is specificity, and adding volume makes the specificity harder to find.

The critique skill does not fix anything. It reviews and writes it up, with six scored dimensions and a prioritised list you can work through. Before any of that it asks whether the piece is even reviewable yet, because reviewing something unfinished produces a list of things you were going to do anyway.

👥

Every review says what is working. Three things, specifically. Not to be nice: without it you cannot tell what to preserve, and the next revision destroys the thing that was already right. That happens more often than anyone admits.

Section 10

The checks that actually run

This is where the pack differs most from anything else you can install, so it is worth a section of its own.

Most design tools give you a checklist and ask the AI to grade its own homework at the end of a long generation, which is exactly the moment it is least able to. Eleven of the twelve skills here carry a working check instead. Real arithmetic, run on your actual file.

ContrastEvery colour pairing measured against the same threshold accessibility regulations use: 4.5 to 1 for body text, 3 to 1 for large text and controls.

Colour visionSimulates the three main types of colour blindness and reports which of your colours stop being distinguishable.

Type and measureWorks out how many characters land on a line, and what column width would fix it if that number is wrong.

Spacing rhythmCounts how many distinct spacing values a piece uses and which sit off the beat.

Image weightReads your actual image files and flags what will make somebody on a phone wait.

VoiceChecks the writing against the habits you recorded, not against a general standard.

Motion safetyReads every duration, flags anything that will read as a jump or a lag, and checks the setting that turns motion off is honoured.

What went missingCompares an old version against a new one and lists every link, contact detail and claim that quietly disappeared.

Take the colour vision one, because it is the clearest example of a check finding something no human eye can.

About 8% of men and 0.5% of women have a red or green colour vision deficiency, which is roughly one man in twelve. The classic failure is a green and a red chosen to look balanced beside each other. Run those through the simulation and they collapse from clearly different into effectively the same colour.

And the balance is what caused it. Matching their lightness so they sit nicely together removed the only channel that would have survived. So the fix the tool gives you is not a different hue, it is a lightness difference, which survives every simulation, greyscale printing, and a bad projector all at once.

💡

Every check separates facts from judgement. Some findings are simply true, like a placeholder left in your file. Others depend on context and are reported with a count so you can decide. A tool that fires on legitimate work teaches you to ignore it, and then it catches nothing at all.

Section 11

The rules it will not break

There is one rule running through all twelve skills, and it is the one worth knowing before you rely on any of this.

⚠️

It will never invent proof. No testimonial nobody gave, no statistic nobody measured, no star rating, no logo strip nobody agreed to, no certification, no before-and-after that is not your work. Not as a placeholder, not "just for now", not greyed out.

The reason is practical rather than moral. Placeholder proof is the most damaging thing a tool like this could produce, because it looks finished. It gets approved, it gets published, and the person publishing it usually does not remember it was a placeholder. Several categories of it are also illegal in most markets.

So where a real value is needed and you have not supplied it, you get a blank in square brackets and a line saying what would fill it. A blank gets noticed and filled. A plausible fake gets published.

The same rule extends to pictures. The pack will happily generate a texture, a pattern, an illustration, or a diagram. It will not generate a photograph of your team, your premises, your van, or your finished work, because a photograph is believed instantly and questioned by nobody.

There are two more rules in the same family worth knowing:

Capture is not correction. Your brand gets recorded as it is, not as it could be improved to be.

A pass is not a redesign. If you ask for a check and the concept turns out to be wrong, you get told so rather than handed a different page. An owner who asked for a check and received a rebuild cannot tell what was wrong with the first one.

Section 12

Improving something you already have

Most people who install a design pack do not want to start from scratch. They have a site, a brochure, or a deck that looks tired.

The refresh skill starts somewhere unexpected: it decides how much should actually change, and the first available answer is leave it.

That verdict exists because the four things most often mistaken for a design problem are not design problems. Nobody is finding it. The offer is unclear. The price is wrong. Or the traffic was never going to buy. A better-looking page seen by the same nobody performs identically, and saying so is the most valuable thing the skill can do.

When change is warranted, it works in order of return, because most people cannot stop and redo everything:

Type → space → colour → structure → content → everything else

The first three cost nothing to reverse and break nothing anyone is relying on. They also reveal what the actual problem was: a great many pieces described as needing a rebuild turn out, once the type is right and the spacing is on a rhythm, to have needed exactly that.

What it protects while it works

A redesign rarely fails because the new version is worse. It fails because something quietly stopped working and nobody noticed for a month. So the skill compares your old version against the new one and reports what disappeared:

Links people saved, bookmarked, printed, or shared. Every one of those now lands on nothing.

Contact details. A number that changed on the page but not on your van is now two numbers, and the wrong one is on the thing you cannot update.

Headings people searched for, and the paragraph that quietly answered the question everyone asks.

🔑

Ask early: what has our current look printed on it? Vans, signs, cards, invoices, packaging. Printed things cannot be redeployed, so either the design accommodates them or somebody is paying to reprint. You want to know which before the work starts.

Section 13

The weekly sweep

The pack installs one background agent, and it exists because of a specific failure mode the pack itself creates.

Remember those blanks in square brackets? They are the honest answer to a missing fact. But they only work if somebody fills them in, and nothing else tracks whether that happened. A blank that never gets filled eventually ships, which is exactly the outcome the rule was protecting you from.

So design-radar sweeps everything the pack has produced, weekly, and catches the four things listed back in Section 03: unfilled blanks, recurring findings, undecided fields the work keeps needing, and finished pieces nobody ever checked.

It reads only your own files and needs nothing connected. You can also call it any time by asking "what still needs doing on my designs".

The recurring-findings check is the one that saves the most irritation. If the pack keeps telling you your colours are wrong across several pieces, that is not three faults. It usually means your brand is missing something your work keeps needing, like a status colour or a second accent, and one upstream fix ends it permanently.

Section 14

A real run, start to finish

Here is what using it actually looks like. A two-person tree surgery outfit adds stump grinding as a new service and wants a page for it.

You: build me a landing page for my new stump grinding service. I want it to bring in work and also get people signed up to my mailing list. Before I build anything, four things...

The direction skill asked which one mattered if only one happened. The answer was the work. The mailing list came off the page, and for a good reason: there was no list and nothing to send, and an email box competes with the phone call for exactly the same reader.

Then the proof inventory. Four stump jobs done in six weeks, photos of all four on a phone, and twenty-three reviews that were all for the tree work rather than the stumps.

That changed the page. The reviews stayed, labelled precisely as tree-work reviews, because implying otherwise is the class of thing that gets businesses in real trouble. The four photographs became the whole case, since they are the strongest evidence available and they were sitting unused. And the count of stump jobs was deliberately left off, because "four" is a true number that argues against you.

The headline that came out:

Eleven years of tree work in [town]. Now we grind the stumps out too.

Sentence case, no exclamation marks, and the word "job" rather than "project", because those were the writing habits captured during setup. Sixteen blanks left in square brackets for the things nobody had supplied yet, including a grinding depth that the tool flatly refused to guess.

And the closing recommendation was not a design change at all: talk through one ordinary job out loud, start to finish, in more detail than feels worth saying. That paragraph is the only part of the page a competitor could not write.

Section 15

Getting the most out of it

Four things that make the difference between a useful install and a great one.

Spend the extra two minutes on setup

Everything downstream reads that file. The anti-references especially: they are checked as constraints on every single thing the pack makes, and they are the cheapest input you will ever give it.

Give it something you actually wrote

The voice capture wants one real message you sent a customer. Not a polished one. That single sample is what makes twelve skills sound like you instead of like a competent stranger, and the voice check measures against your habits rather than against a general standard.

Fill the blanks in

When you see square brackets, that is the pack refusing to make something up. Fill them and the piece is finished. Ignore them and the weekly sweep will keep reminding you, which is exactly what it is for.

Connect a design tool if you have one

Nothing needs connecting and everything works without it. But if you keep a brand kit or a token set somewhere, connecting it turns a five-minute capture into a thirty-second one and removes every transcription error. Browsing helps too, because several checks are impossible from a file: real contrast where text sits over a photograph, what happens at a phone width, and what a page said before you replaced it.

🎯

The one thing to take away: run the substitution test on the next thing you make. Take your name and colours off, put a competitor's on, and see what else would have to change. If the answer is nothing, you now know exactly what to do about it.

Most design advice tells you what to avoid. That produces work that is clean, compliant, and completely inert, because a list of prohibitions tells you what not to be and gives you nothing to become.

The thing that actually makes work yours is specificity, and the strongest source of it is the detail you think is boring. What you consider unremarkable about how you work is very often the only thing a customer would find distinguishing, precisely because you have stopped noticing it.

Start with "set up my brand". Everything else follows from there.