Taste

"I like it" is preference. "This works because the hierarchy guides your eye to the one action that matters" is taste. The difference is that taste comes with reasons — reasons grounded in what the thing is for, who it serves, and the accumulated knowledge of what has worked on humans for centuries. Preference is private and arbitrary; taste is arguable, teachable, and therefore valuable.

This reframing matters because it makes taste trainable. You cannot practice having better feelings. You can absolutely practice connecting what you feel to why — and that loop, run thousands of times, is the entire education.

Practice. Pick one product you love and one you find ugly. For each, write five sentences of reasons — no "I like," no "it feels." Every claim must reference something concrete on the screen.

Early in any craft there is a painful gap: your taste is good enough to know your own work is not. Most people resolve the pain by lowering their taste. The ones who get good resolve it with volume — making enough things that the hand catches up to the eye. AI changes the economics of this completely: the volume that took a decade now takes a season, because each attempt costs minutes instead of days.

The gap never fully closes; it moves up. That is what improvement feels like from the inside — perpetual mild dissatisfaction. Calibrate to it. The person with no gap left is not a master; they have just stopped seeing.

Practice. Make the same small artifact — a poster, a landing hero, an app screen — ten times in one week with AI assistance. Rank all ten. Write down what the top one has that the bottom one lacks.

Your eye is a model, and it is trained on whatever you feed it. Feed it the algorithmic mid-stream of whatever scrolls past, and your taste converges to the average. Feed it deliberately — the best typography of the last century, the best interfaces shipping today, architecture, film stills, museum paintings — and your defaults quietly rise. Every person whose taste you admire has a private museum in their head. They built it on purpose.

Collect actively, not passively: save what stops you, and write one line about why it stopped you. The library matters less than the noticing.

Practice. Start a swipe file today. Add five excellent things per week, each with one sentence on why it earns its place. Review the file monthly and evict anything that no longer impresses you.

Most looking is recognition — a tenth of a second of "landing page," and the mind moves on. Seeing is slower: where does the eye land first, and was that chosen? What is aligned to what? How many font sizes are in play, and do the differences mean anything? Why does this corner feel crowded? Designers are not people with magic eyes; they are people who have practiced looking at one thing for ten minutes without getting bored.

The slow look is also your quality-control instrument for AI output, which is engineered to survive the fast look. Plausible at a glance, sloppy under attention — that is the signature of generated mediocrity.

Practice. Choose one screen you use daily and look at it for ten full minutes. Write down twenty observations — things that are actually there, not judgments. The last five will be the interesting ones.

Critique has a grammar: start from intent ("what is this trying to do?"), judge against the intent rather than your preferences, be specific enough that the maker knows what to change, and separate the work from the person. "Make it pop" is not critique; "the primary action competes with three other elements of equal weight" is. Learning this language does double duty in 2026 — it is exactly the language that steers AI tools, which respond to precision and shrug at vibes.

Receiving critique is the same skill mirrored: extract the observation from the delivery, thank the observer, and decide separately what to act on. Defensiveness is the tax that stops people from giving you the good stuff.

Practice. Critique one piece of work this week — a colleague's, a famous product's, your own — using the grammar: intent, what serves it, what fights it, one concrete suggestion. Four sentences minimum, zero adjectives of pure preference.

Original-from-nothing is a myth amateurs believe and professionals have abandoned. Real work starts from references: three or four existing things that each solve part of your problem, studied closely enough to understand why they work, then recombined for your specific context. The theft is in copying the surface; the craft is in extracting the principle. This is also the single highest-leverage way to direct AI tools — a reference communicates a hundred decisions that a paragraph of prompt cannot.

Build the habit of asking, before any project: who has solved a problem shaped like this one, brilliantly, possibly in a completely different field?

Practice. For your current project, assemble a reference board of four works. Annotate each: the principle it demonstrates and the part of your problem it informs. Use the board to brief an AI tool and compare against your unbriefed result.

Most design is typography, and most typography is hierarchy: making the order of importance visible so the reader never has to work out where to look. The core moves are few — size, weight, spacing, and restraint. One or two typefaces. A deliberate scale instead of improvised sizes. Real differences instead of timid ones: if two levels differ, make them obviously differ. Line lengths a human can scan, line heights that let text breathe.

Test any layout by squinting: the structure that survives blur is the structure the user actually perceives. If everything is equally loud, you have said nothing.

Practice. Take a real page of yours and rebuild it using only one typeface, three sizes, and two weights. No color, no boxes. Make the hierarchy carry everything. Compare with the original — honestly.

Amateurs treat empty space as waste to be filled; professionals treat it as the most expressive material they have. Space is how you group without boxes, separate without lines, and confer importance without size. The working rules: related things sit close, unrelated things sit measurably farther apart, and the spacing system repeats — the same handful of gaps everywhere, never a new value invented on the spot. Rhythm comes from that repetition, calm comes from rhythm.

Generosity reads as confidence. Cramped layouts whisper that nobody decided what mattered, so everything was kept. The hard part of whitespace was never aesthetic; it is the decision about what to cut.

Practice. Audit one of your layouts and list every distinct spacing value in it. Reduce the list to four values on a consistent scale and rebuild. Watch what the discipline does to the feeling of the page.

Color is the fastest channel to emotion and the easiest place to lose control. The professional defaults are unglamorous: mostly neutrals, one color that means something (action, brand, attention), used so consistently that its appearance is information. Saturation is volume — a page where every element shouts in full saturation says nothing, loudly. And contrast is not a stylistic preference; it is whether people can read your work at all, in sunlight, with aging eyes.

Steal palettes from the physical world — paintings, landscapes, old book covers — rather than from color pickers. Nature's combinations come pre-balanced in ways hex-code roulette never is.

Practice. Reduce a screen of yours to neutrals plus exactly one accent color, applied only where you want the eye to go. Run the contrast checks. Note what the constraint clarified.

The AI-native design loop inverts the old economics: exploration is free, so explore properly. Generate twenty genuinely different directions before lunch — different structures and different moods, not twenty variants of one idea. Then switch roles completely: from generator to curator. Kill eighteen without mercy, and push the surviving two through another wide round. Divergence, then ruthless convergence, repeated.

The failure mode is settling for the first competent option — and AI's first option is always competent. Competent is the new baseline, which means competent is the new invisible.

Practice. For one real design task, force twenty meaningfully distinct AI-generated directions. Select two, articulate in writing why those two, and iterate only on them. Compare the result to your usual first-idea workflow.

Prompting asks the machine to guess your taste; directing supplies it. The difference is a creative brief — the same instrument design leads have used on studios for a century: who this is for, what one feeling it must produce, three reference works with the principle each contributes, and the things that are absolutely off the table. With that brief, AI tools become a remarkably good studio. Without it, they return the statistical average of everything, which is precisely what mediocrity is.

Critique the output in the language of Module 2 and feed the critique back. The tools take direction better than most juniors — but only from someone who can give it.

Practice. Write a one-page brief for a real task: audience, the single feeling, three annotated references, the forbidden list. Run it through your AI tool of choice. Then run the same task with your old two-line prompt and put the results side by side.

Taste tells you which direction is promising; only contact with reality tells you if it works. AI collapsed the cost of finding out — a clickable, working prototype in an afternoon — which removes the last excuse for debating mockups in meetings. Put the real thing in five users' hands and watch. Where they hesitate is your hierarchy failing. What they don't notice is your emphasis failing. What they ask about is your copy failing.

Taste plus testing is the complete loop: judgment proposes, reality disposes, and every cycle sharpens the judgment for the next proposal. Designers who skip the second half are decorating; designers who skip the first half are flailing.

Practice. Build a working prototype of a current idea in one day using AI tools. Watch five real people use it — silently. Write down the three places reality disagreed with your taste, and what that teaches your eye.

Almost everything is improved by removal: the third typeface, the second metaphor, the feature nobody asked for, the section that exists because someone senior wrote it. Addition feels like progress and subtraction feels like loss, which is exactly why disciplined removal is rare enough to be a signature. The question that powers it: "if this element vanished, would anyone's experience get worse?" Be honest, and most elements fail it.

AI makes this discipline urgent — generated work trends toward more: more words, more effects, more sections, more everything. The human contribution, increasingly, is deciding what does not survive.

Practice. Take a finished piece of your work and remove 30% — of the words, elements, or features — without losing anything essential. If you can't find 30%, you are not looking; ask an AI to argue for each cut, then judge.

"Done" is a taste judgment, and both of its failure modes are expensive. Shipping early leaves the rough edge every user feels and nobody reports. Polishing forever hides fear of judgment behind craftsmanship. The professional's test: the work is done when remaining changes are preferences rather than improvements — when you are no longer fixing, just rearranging. That moment is feelable, but only if you pause and ask.

A practical ritual: before shipping, one full pass through the work as a stranger — fresh eyes, real device, no creator's autopilot. Fix what genuinely snags. Ignore what is merely different from how you'd do it today. Ship.

Practice. On your next piece of work, write the definition of done before you start: the three things it must achieve. When they're met, do one stranger's pass and ship — even though you can see five more tweaks. Especially because you can.

One beautiful screen is taste; a hundred coherent screens is a system. Design systems — the named colors, the spacing scale, the component library, the voice guidelines — are how quality survives contact with deadlines, teammates, and machine generation. Every decision encoded in the system is a decision nobody re-makes badly at midnight. This matters double with AI in the pipeline: a tool briefed with your system produces your product; a tool without one produces generic output wearing your logo.

Systems are edited, not just built. The discipline is refusing additions — every new component, color, or pattern must justify why an existing one cannot serve.

Practice. Write the one-page version of your design system: typefaces, type scale, spacing values, colors with their meanings, and five rules of voice. Brief an AI tool with it and generate a new screen. Score the consistency against your real product.

When everyone can produce competent work instantly, competent work is worth nothing — the price of any abundant commodity. Attention, meanwhile, stays as scarce as ever. The economic consequence is brutal and wonderful: the premium on the top 1% of quality goes up as the cost of the bottom 99% goes to zero. Markets where production gets cheap don't level out; they polarize into a vast gray sea and a few things people actually love.

This is why taste is a 2026 career strategy, not an aesthetic indulgence. The sea is machine-made. Being reliably on the right side of the polarization is the part machines cannot do alone — they average; you must choose.

Practice. In your own market, name the gray sea (what does baseline AI-generated competence look like?) and the three products that float above it. For each of the three, identify the specific quality decisions that lift it. Those decisions are your map.

"Feel" is what users call the accumulation of decisions too small to list: the interface responds instantly, the error message says what to actually do, the empty state teaches instead of apologizing, the animation confirms rather than performs. No single one matters. Together they are the difference between software people use and software people recommend. Feel cannot be added in a polish sprint — it is a standard applied to ten thousand small calls as they happen.

This is taste's operational form: not the big redesign, but the daily refusal to ship the small wrongness. Teams have the feel their leader's tolerance allows.

Practice. Spend thirty minutes in your product hunting only small wrongnesses: the misaligned row, the jargon error, the 800ms pause. List fifteen. Fix five this week — with AI doing the labor, there is no excuse left.

A brand is not a logo; it is the sum of every choice a company has made, as perceived by people deciding whether to trust it. Consistent taste, applied for years, compounds into the most defensible asset in an AI economy — because it is the one thing that cannot be generated on demand. A competitor can clone your features in a weekend and your visual style in an afternoon. They cannot clone a decade of choices that all pointed the same direction.

The practical discipline: know what your brand refuses. Aesthetic positions are defined by their exclusions — the colors you'll never use, the tone you'll never take, the dark pattern you'll never ship. What you refuse is what people learn to trust.

Practice. Write your brand's refusal list — ten things it will never do, visual and behavioral. Test it against your last month of output. Every violation you find is either a mistake to fix or evidence the list is aspirational fiction; decide which.