An AI designer,
available right now.

The reason design reviews go badly is that without a stated goal, every comment is a preference and the most senior preference wins. Nobody learns anything and the work does not improve.

The Design Copilot starts from what the screen is trying to achieve, which converts "I do not like the button" into "the primary action is competing with three others" - a claim that can be evaluated.

Almost every product that looks inconsistent looks that way because decisions were made one screen at a time. Twelve type sizes, five greys, spacing values that nobody chose.

A type scale and a spacing scale are an afternoon's work and they prevent the slow accumulation that takes a quarter to unwind. This is the least glamorous and highest-leverage thing in product design.

Contrast ratios, focus states, target sizes and text alternatives are cheap at design time and expensive as a retrofit - particularly contrast, which usually means renegotiating a palette that has already been approved.

WCAG AA is the working standard: 4.5:1 for body text, 3:1 for large text and interface components. Checking a palette against it before anything is built is a five-minute step that saves a fortnight.

The complaint that something feels off is almost always a hierarchy problem, and hierarchy is one of the few areas of design where the rules are concrete enough to check.

A layout has a hierarchy whether or not one was designed. The eye goes somewhere first, and if that place is not what the screen is for, the screen is working against itself. Two elements competing for primary attention means neither has it.

The tools that create hierarchy are size, weight, colour, contrast, and space - and space is the one that gets used least and works best. Elements grouped tightly read as related; elements separated read as distinct. Most cluttered interfaces are not carrying too much content, they are carrying content with undifferentiated spacing, so nothing groups.

Contrast is the second underused tool. A page where everything is bold has no emphasis. Making one thing prominent usually means making the other things less so, which feels like a loss and is how emphasis works.

The practical test is to squint at the screen or blur the screenshot. What remains legible is the hierarchy as a user experiences it in the first second. If the primary action is not what survives, that is the problem, and it is a specific fixable one rather than a matter of taste.

A design system fails in a predictable way: it is built thoroughly, adopted partially, and then bypassed under deadline until it describes a product that no longer exists.

The systems that survive are the ones that are easier to follow than to ignore. That means fewer decisions, not more documentation. A type scale with five sizes gets used; one with twelve is a menu, and a menu recreates the inconsistency it was meant to solve.

The same applies to spacing. A scale based on a consistent step - four or eight pixels, multiplied - removes the per-decision judgement entirely. Any value not on the scale becomes visibly a choice, which is exactly the friction you want.

Colour needs roles rather than names. A palette of hex values invites arbitrary use. A set of semantic tokens - surface, ink, accent, border, and their states - encodes intent, which means a theme change is a token change rather than a search across the codebase.

And components need to cover the awkward states. A button component with default, hover, active, focus, disabled and loading is a component. One with default and hover is a suggestion, and the missing states get invented separately by everyone who needs them.

Formal user research is expensive and often skipped entirely, which leaves teams designing on assumption. There is a substantial middle ground that costs almost nothing.

Watching five people attempt a task is the classic example, and the finding that a handful of participants surfaces the majority of usability problems has held up well in practice. It does not need a lab, a script, or a research function - it needs five people, one task, and the discipline not to help them.

Not helping is the hard part. The instinct when someone gets stuck is to explain, and the moment of getting stuck is the entire data. Watching in silence is uncomfortable and it is where the value is.

Support tickets are a research corpus that already exists. Recurring confusion in support is a design problem that has already been reported, categorised and paid for, and almost nobody in design reads them.

Analytics tells you where people stop and never why, which makes it a good instrument for choosing what to investigate and a poor one for deciding what to change. Pairing a drop-off number with five people attempting that step is the combination that actually answers a question.

Designs are made in the ideal case - real content, sensible length, everything loaded, nothing wrong. Products spend a meaningful share of their time in none of those conditions.

Empty states are the first gap. A dashboard with no data, a list with no items, a search with no results. These are the screens a new user sees first, which makes them the highest-stakes screens in the product and the ones designed last. An empty state that explains what goes here and offers the action to create it is onboarding; a blank panel is confusion.

Loading states are the second. Something has to occupy the gap, and a layout that shifts when content arrives is worse than one that reserves the space. Skeletons work because they set the expectation of shape.

Error states are the third and most neglected. An error message that names what went wrong and what to do next is a design artefact. One that says something went wrong is an admission that nobody thought about it.

Then the extremes of real content. A name three times longer than the placeholder, a list with four hundred items, a title with no spaces. Every one of those exists in production and none of them exist in the mockup.

Designing the awkward states is unglamorous and it is most of the difference between a product that demos well and one that holds up.