Interface work that holds up: accessibility, restraint, feedback and speed

Design advice arrives as a list of trends with a year attached to it, and it ages exactly as fast as the title implies. Some of what’s on those lists is genuinely load-bearing and will still be true in a decade. Some is decoration that will look dated before the project ships.

This is an attempt to separate them, from the position of people who have to build the result and then keep it working.

Accessibility is the constraint everything else sits inside

Start here, because it’s the one thing on the perennial trend list that isn’t a trend, and it’s the one most often treated as a late-stage checklist.

The Web Content Accessibility Guidelines are the reference standard, and for a US audience they matter for reasons beyond principle: accessibility shows up in enterprise and public-sector procurement questionnaires, and it is a live area of litigation for consumer-facing sites. If you sell to large organisations, someone will eventually ask you which conformance level you target. Having an answer is cheaper than acquiring one under deadline.

What it actually asks for is unglamorous and mostly decided during design rather than during build:

• Contrast that survives the real world. Text against its background, and (the one people miss) interface elements and focus indicators too. Light grey placeholder text on white is the single most common failure and it fails for everybody outdoors.
• Type you can read and resize. Legible faces, sensible line length and height, and a layout that doesn’t collapse when someone increases the browser’s text size.
• Everything reachable by keyboard, in a sensible order, with a visible focus state. If you cannot tab through your own checkout, neither can a screen-reader user.
• Labels on form fields, and errors that say what to do. “Invalid input” is not an error message. And a placeholder is not a label; it disappears the moment someone starts typing, which is exactly when they need it.
• Alternative text that carries the information, not the filename.
• Captions and transcripts on video, which are also how most people watch video anyway.
• Meaning never carried by colour alone. A red field border with no text is invisible to a large number of people.

Two things worth saying plainly. First, almost every item on that list also makes the product better for people with no impairment at all: on a poor connection, on an old device, in bright sun, one-handed, distracted. Accessibility work and mobile-robustness work are largely the same work with two different justifications.

Second, retrofitting it is a real project and building it in is close to free. It’s the same economics as right-to-left support, which we’ve written about separately: the cost is proportional to how many decisions you have to go back and undo.

Functional minimalism, and how it gets misused

Restraint in an interface isn’t an aesthetic preference. It’s the absence of things competing with the action you want someone to take. A page with one obvious next step outperforms a beautiful one with four equally weighted options, and that has nothing to do with taste.

The misuse is treating minimalism as a licence to remove information. Icon-only navigation with no labels, invisible affordances, controls that only appear on hover, form fields with the label stripped out; these are all “cleaner” and all worse, and several of them break accessibility outright. Whitespace is free. Clarity that someone has to hunt for isn’t.

The test is not whether the screen looks calm. It’s whether a first-time user can name what this page is for and what they’re supposed to do next without being told.

Microinteractions: feedback, not decoration

Small responses to user action (a button state change, a subtle confirmation, a progress indicator) do a specific job: they tell someone the system received the input. That matters most when something takes a moment, because in the absence of feedback people assume failure and click again, and the second click is where duplicate orders come from.

The useful version of this is a rule about states rather than a rule about animation. Every meaningful screen has at least five: default, empty, loading, error, and success. Most design deliverables contain the first one. The other four get invented during development, under time pressure, by whoever is closest, which is why the empty state of your app is usually a blank rectangle and the error state is a raw message from an API.

Design the empty state especially. It is the first thing every new user sees, and it is the only screen where you have their full attention.

On motion: keep it short, keep it purposeful, and honour the operating system’s reduced-motion setting. Animation that can’t be turned off makes some people physically unwell, and the flourish is never worth it.

Dark mode is a second product, not a toggle

Offering a dark theme is now close to expected, and it’s frequently underestimated. It is not an inversion. Contrast ratios have to be re-checked in the second theme, pure black backgrounds with pure white text produce eye strain rather than relieving it, shadows stop working as a depth cue, brand colours that sing on white often go muddy or fluorescent, and every image, illustration, chart and PDF export needs a decision.

The honest position: it doubles the visual QA surface. Do it deliberately, with tokens rather than one-off colour values, or don’t do it. A half-implemented dark mode with three unreadable screens is worse than a light-only product.

Speed is a design decision

Performance gets filed under engineering, and most of it is decided by design: how many images, at what resolution, how many typefaces and weights, how much motion, how many third-party widgets, whether the hero is a video.

This matters commercially in two directions: people leave slow pages, and search ranking accounts for real-world loading behaviour. And it matters most on mobile, on a mid-range device, on a connection that isn’t yours. Test there rather than on the machine the design was made on.

A redesign that is slower than the thing it replaced is a downgrade, however much better it looks in a case-study screenshot. Measure before and after as part of the project, not as a favour afterwards.

Where visual identity genuinely earns its keep

There’s a version of the design argument that overclaims (that better-looking design increases conversions), and it’s worth correcting, because the real mechanism is more useful than the exaggerated one.

Visual design does three things reliably. It buys attention in an environment where nobody has committed any yet. It carries information faster than prose: a diagram, a chart or a short explanatory video will land a structure that three paragraphs won’t. And it signals seriousness, which is a proxy people use, fairly or not, for whether a company will still exist next year and whether it can be trusted with a payment.

What it can’t do is fix an unclear offer, a broken process or a product nobody wants. A polished interface on a confusing proposition makes the confusion more legible, not less.

The practical form of this is consistency, and consistency at scale means a design system rather than a style guide: tokens for colour, type and spacing; a component library shared between design and code; and states defined once rather than per screen. A PDF describing the brand is a document. A component library is infrastructure, and it’s the only version that survives the second team touching the product.

Where design meets the build

The gap between a design file and a shipped interface is where most quality is lost, and it’s almost always the same handful of things.

Design with real content. Real product names, the longest one you have, real prices, a customer with a hyphenated surname, a category with two items and a category with four hundred. Placeholder text hides every layout problem you’re about to have.

Design for the languages you’ll ship. Translated strings are commonly much longer than English ones, and a button sized to its label breaks in the first locale. If a right-to-left market is anywhere on the roadmap, the decision to make today is simply to stop using physical directional values; the full argument is in our piece on right-to-left interfaces.

Decide what’s a component and what’s a one-off, explicitly, before the build. Every undeclared one-off becomes a maintenance obligation nobody scheduled.

Hand over behaviour, not just appearance. What happens on tap-and-hold, what happens offline, what the validation rules are, what the maximum length is. Anything unspecified gets specified by a developer at speed.

AI in the design toolchain

Generative tools now produce competent layouts, imagery and copy variants, and they’ve genuinely compressed the exploration stage: more directions, faster, at the point where you’re still deciding what the thing is.

They are weakest at exactly the things above. They don’t know your real content, your edge cases, your accessibility target, your five states, or that the third menu item is a legal requirement. Used to explore, they’re a real accelerant. Used to produce final interface work unreviewed, they generate confident, generic, and frequently inaccessible screens, the same failure mode as everywhere else in this business, which is a plausible output nobody checked.

What to ask before signing off a design

• What conformance level are we targeting, and who checks it? Before build, not after.
• Show me the empty, loading and error states. For every screen that has them.
• Show it on a mid-range phone on a bad connection. Not a desktop mockup.
• Is this a component or a one-off? For every element that appears more than once.
• What does this look like with our real worst-case content? Longest name, most options, no image.

Get those answered and the trend questions mostly answer themselves, because almost everything on the list either survives them or turns out to have been decoration.

We design and build the interface and the system behind it in the same team, which mostly means the states, the real content and the accessibility questions come up while they’re still cheap to answer. That’s included app store products with text-to-speech and gamification (Walnutflow), AR experiences (Droyz Toys), a booking platform where availability is the product (MyGlampings) and site migrations where the existing calls to action and measurement were preserved deliberately rather than rebuilt from memory (Deren). If you have a design that looks right and behaves badly, send us the screens you’re least happy with.