The brief is not the problem
Clients rarely misdescribe what they want. They misdescribe why they want it — and the ‘why’ is the sentence worth spending a week rewriting, because everything downstream inherits it.
Notes on practice, written between projects.
Nine essays about how the work actually goes: rewriting briefs, defending weight budgets, and the parts of design practice nobody puts in a case study.
- 9 essays
- Roughly monthly since 2025
- No guest posts
Latest
Latest essay
The archive
All nine.
Filter by subject, then choose one — every essay is published in full further down this page.
The brief is not the problem
Why the first week of every engagement is an argument, and why that is the cheapest week you will buy.
Designing for the first ninety seconds
Onboarding is not a funnel. It is the only time anyone will read what you wrote.
What a design system owes its engineers
A library that is pleasant to look at and unpleasant to use is a liability with a good screenshot.
Colour under pressure
Accessible palettes that still have a point of view, and the maths that gets you there.
We stopped running discovery workshops
Twenty people, sticky notes, and a consensus nobody actually held. What replaced it.
Motion is a load-bearing wall
Animation is not decoration. It is how an interface explains what just happened.
Typography for interfaces that get read
Most interface type is designed to be admired at 200% zoom in a portfolio shot.
The case against the homepage carousel
Yes, again. The data has not changed and neither has the argument.
Measuring brand when the funnel cannot see it
Attribution models are honest about what they can count and silent about the rest.
Nothing filed under that subject yet.
In full
The essays.
The brief is not the problem
Clients rarely misdescribe what they want. They misdescribe why they want it, and every downstream decision inherits that error.
A brief arrives asking for a website. Underneath it, nine times out of ten, is a sales team that cannot explain the product, or a pricing page that contradicts the pitch deck, or three product lines that were acquired rather than designed. Build the website and none of that moves. Build it well and you have made an expensive, beautiful restatement of the confusion.
So the first two weeks of every engagement here are spent trying to break the brief. We interview the people who wrote it and the people who will have to live with it, and we ask the same question in six different ways: what would have to be true for this project to be unnecessary? The answers are usually more interesting than the brief.
This is not a clever consulting move. It is a cost-control measure. A wrong brief discovered in week two costs a fortnight. The same wrong brief discovered in month five costs the project, and often the relationship. Clients almost never object once you frame it that way, and the ones who do object are telling you something useful about how the engagement is going to go.
The output is a rewritten brief, in your words, with the trade-offs made explicit. Sometimes it is a smaller project than the one you asked for. Twice in eleven years it has been no project at all, and both times we said so and sent an invoice for the fortnight.
Designing for the first ninety seconds
Onboarding is not a funnel. It is the only ninety seconds in the entire relationship when someone will read what you wrote.
Watch a first-time user and the pattern is always the same. There is a short window — roughly a minute and a half — of genuine attention, in which they are willing to read a sentence, make a decision, and be wrong about something. After that the reading stops permanently. Everything from then on is pattern-matching against interfaces they already know.
Most products spend that window on a product tour. Four screens of features, a progress dots row, and a Skip link in grey. The window closes, and the first real decision the person makes is one they now make blind.
The better use is a single meaningful choice with real consequences. What are you here for? What should this show you first? Put the tour behind that. The difference in downstream behaviour is not subtle: on Solstice Health, moving the one real question to the front and deleting three explanatory screens took median completion from 4m 20s to 72 seconds.
The uncomfortable corollary is that most onboarding copy should be deleted rather than improved. If a sentence is not helping someone make the first decision, it is spending the only attention you will ever be given.
What a design system owes its engineers
A component library that is pleasant to look at and unpleasant to use is a liability with a very good screenshot.
Design systems are usually judged by their documentation site. They should be judged by how often engineers work around them. That number is knowable: count the one-off components in the codebase, six months after launch. If it is climbing, the system is failing regardless of how the tokens page looks.
Three things drive that number down. First, every component needs an escape hatch — a documented, blessed way to do the thing it does not support, so the workaround is inside the system rather than beside it. Second, the contribution path has to be shorter than rebuilding. If adding a variant takes three meetings, nobody adds a variant. Third, deprecation needs a policy with dates, otherwise nothing is ever removed and the library becomes an archive.
None of this is design work in the usual sense, which is exactly why it gets skipped. It is also the difference between a system that is still in use in year three and one that quietly forked in month eight.
On Kestrel Mobility we spent nearly as long on the contribution and deprecation rules as on the components. Eighteen months later the library had not fragmented across six regional teams. That is the whole result.
Colour under pressure
Accessible palettes that still have a point of view. The constraint is real, and it is not the reason your colours are boring.
The complaint is familiar: contrast requirements flatten a palette into corporate blue and grey. It is not true, and the belief costs studios their best colour work. What flattens a palette is picking hues first and checking contrast afterwards, then desaturating everything until it passes.
Work the other way. Fix the luminance relationships first — decide what is text, what is surface, what is accent, and at what contrast ratio — then choose hue and chroma freely within those bands. A colour space with perceptual lightness makes this practical: you can rotate hue and adjust chroma at a fixed lightness and the contrast ratio barely moves.
The second thing that helps is being honest about where a colour is allowed to appear. A saturated accent that fails on body text can be perfectly legitimate on a 3px rule, a large numeral or a hover fill. Palettes get dull when every colour has to survive every context. Give each one a job.
The result is usually more distinctive, not less. Constraints do not remove the interesting decisions; they remove the arbitrary ones, which is what makes a palette look designed rather than sampled.
We stopped running discovery workshops
Twenty people in a room, four hours, eleven grids of sticky notes, and a consensus that nobody actually held.
The workshop format produces a specific failure. The loudest three people set the frame in the first twenty minutes. Everyone else calibrates to it. The output looks like agreement and is really a record of who spoke first, which you then take away and design against for four months.
We replaced it with individual interviews: forty-five minutes each, same eight questions, no group. It takes longer — two weeks instead of an afternoon — and it produces something a workshop structurally cannot, which is disagreement. When four people describe the customer four different ways, that is the actual finding. In a workshop it never surfaces.
We still get everyone in a room, but afterwards, and for a different purpose: to show the disagreement back and force a decision on it. That meeting is short, uncomfortable and worth more than the workshop it replaced.
The only real cost is that it looks like less. There is no wall of sticky notes to photograph. Clients who need the theatre of collaboration find it unsatisfying, and we have learned to say that out loud before we start.
Motion is a load-bearing wall
Animation is not decoration. It is how an interface explains what just happened and where the thing you were looking at went.
Remove every transition from a product and it does not become cleaner, it becomes harder to follow. Panels appear from nowhere. Deleted rows vanish without trace. The user rebuilds the model in their head each time, and the cost of that shows up as hesitation rather than as complaints.
Which is why motion deserves tokens, not vibes. Durations quantised to a short scale, easings assigned by intent — entering, exiting, moving, expanding — and choreography rules that say what happens when three things change at once. Written down, in the same document as the type scale.
The reduced-motion path is part of the design, not a switch that turns everything off. Someone who has asked for less motion still needs to know that the row was deleted. Usually that means replacing movement with a fast cross-fade and keeping the timing, so the rhythm of the interface survives.
And the counterweight: every animation should be deletable. If removing it costs nothing, it was decoration, and we take it out. That test kills roughly half of what gets proposed here, including plenty of mine.
Typography for interfaces that get read
Most interface typography is designed to be admired at 200% zoom in a portfolio shot, then shipped at 14px to someone squinting on a train.
Two habits cause most of the damage. The first is choosing a display face for body text because it looks characterful in a specimen. At paragraph size, on a mid-range Android, with subpixel rendering that flatters nothing, that character becomes noise. The second is a type scale with too many steps — nine sizes where four would do — which produces hierarchy nobody can perceive and a maintenance burden everybody can feel.
What actually helps: measure. Set the body size from the line length you can achieve at your container width, not the other way round. Sixty to seventy-five characters, then choose the size that lands there. Everything else follows.
Then use weight and space before you use size. A 15px label at 600 with generous space above reads as a heading perfectly well, and it keeps the vertical rhythm intact on a small screen where a 32px heading would eat a third of the viewport.
Save the display face for the two or three moments that deserve it. Scarcity is what makes it feel like a decision rather than a default.
The case against the homepage carousel
Yes, again. The data has not changed, the argument has not changed, and the carousel is still on the wireframe.
The carousel is not a design pattern. It is a conflict-resolution device. Five teams want the hero slot, nobody will decide, so everyone gets four seconds. The cost is paid entirely by the visitor, who sees slide one, does not wait for slide two, and forms an impression from a message chosen by rotation order.
The honest version of the conversation is: which of these five is the reason someone should keep reading? That question is hard, political and exactly the one the project is meant to answer. The carousel exists to avoid asking it.
If the answer genuinely is ‘it depends who is looking’, that is a segmentation problem with real solutions — distinct entry points, a paid-media landing set, or personalisation with an actual signal behind it. All of them are better than a rotation.
We have lost work over this. Not much, and never a project we would have enjoyed.
Measuring brand when the funnel cannot see it
Attribution models are honest about what they can count. The trouble starts when we treat what they cannot count as zero.
Brand work moves things that show up as absences: the deal that closed faster, the enquiry that arrived already convinced, the candidate who applied without a recruiter. None of them appear as a line in an attribution report, which is why brand budgets lose arguments to channels that can produce a number by Friday.
The fix is not a better model. It is measuring the things brand actually changes, and baselining them before you start. Four we use routinely: sales-cycle length by segment; the proportion of enquiries that arrive with a budget already stated; unprompted recall in a small quarterly panel; and cost per qualified applicant on hiring.
They are all lagging, all noisy, and all considerably more honest than a last-click attribution of a rebrand. On Orrery Labs the number that moved was sales-cycle length — down thirty-five per cent — because the catalogue finally explained itself before a salesperson had to.
The discipline is to write down which of these you expect to move, and by roughly how much, before the work starts. Then you have a claim that can be wrong, which is the only kind worth making.
Disagree?
Tell us where we are wrong.
Genuinely — the best arguments we have had started as a reply to one of these. Or subscribe below and get the next one first.