Taste Is
Not a Gift.

For most of my career, the hard part was making the thing: writing it, designing it, shipping it. I spent over a decade getting better at that. Then in about eighteen months, the constraint disappeared. What's left is the question that used to come last in a project: should this exist at all? That used to be a bonus skill for senior people. Now it's everyone's full-time job.

The strange part is how fast the muscle most of us built stopped mattering. We're all still adjusting.

Taste isn't a feeling about what's pretty. It's the everyday work of choosing one version of a thing over eleven others, and being right enough, often enough, that the people around you stop second-guessing you. Anu Atluru wrote a line I keep coming back to: in a world of scarcity, we treasure tools; in a world of abundance, we treasure taste.

The thing that used to be a luxury is now the job.

A model can give you a thousand variants of anything. What it can't do is tell you which one is worth keeping. Whatever "view" it has is a smoothed average of its training data: confident, fluent, and frozen at the moment the data was collected. It can make a probabilistic guess about what you'd want. It can't form a real point of view about what's worth making, because that requires being in the world right now, with these people, in this moment.

That gap, between what's possible and what's worth making, is where human judgment now lives. The teams that win in the next five years will be defined less by what they built and more by what they chose not to build.

Rick Rubin said it cleanest: "The reason we go to the artists we love is for their point of view. The AI doesn't have one."

Teams ship more now. They ship faster. The review cycles I used to count on are gone. A leader who can only exercise their own taste, and can't help it spread to other people, becomes a bottleneck the team works around. Or worse: a ghost haunting the product with decisions they didn't actually make.

The work, I think, is no longer to have good taste. It's to make your taste useful to other people.

Most of us have one or two of these and call it taste. Myself included, until I started really looking at where my own decisions came from. The four ingredients are what you need together: conviction, reps, empathy, and context. When all four are there, the work feels right. When one is missing, you can usually name which one. That's the whole framework.

The argument we have in the meeting after a launch is almost always about which one of these was missing, or needed more of.

In between sit five layers: perception, decision, craft, system, transmission. Most projects I've led get past the first two and stop. The hard work is the last three: turning what one person knows how to do into something the whole team can do without them in the room.

Companies don't lose their taste in one quarter. They lose it slowly, across many small failures of system and handoff.

It's committing to a direction before the data backs you up, and holding the direction when the criticism comes. Without it, teams hedge. They A/B test things that should have been one person's call. The product ends up reading like the average of every opinion that passed through it.

An example. A team is building an AI writing assistant. Half the team thinks the assistant should be invisible, auto-completing in the background, fading into the editor. The other half thinks it should be a visible companion. A sidebar with personality, suggestions you can talk to. The data is split. The user research could be read either way. At some point, someone has to choose, knowing they might be wrong, knowing the choice will shape the next two years of work. The teams that ship something distinct are the teams where someone does. The teams that ship something forgettable are the teams that try to do both.

I've watched "let's see what the data says" become the polite way to defer a decision no one wants to make.

It's not built by reading about taste. It's built by looking carefully at lots of work, slowly, with stillness, and being honest about what's actually working in each one. Which part of this onboarding flow makes you feel welcomed? Which part of that pricing page makes you trust it? The looking is the practice.

An example. A designer wants to redesign a notification system. The "right" way (the one most teams default to) is to A/B test six variants over four weeks. The taste-led way is faster and looser: pull six notification systems you respect, like Linear, Notion, Slack, Intercom, Stripe, Things, and study what each one is doing. Pick the one detail from each that resonates. Sketch a version that combines the four or five you find most truthful. Ship it to a small group. Learn. Adjust. The first version might not be right. That's fine. The point is the loop: look closely, decide, ship, look again. A team that runs this loop fifty times in a quarter builds taste twice as fast as the team A/B testing eight things to statistical significance.

The throwing-away is where the taste lives. We had to learn that the hard way.

Taste without an audience is just preference, served with confidence. The fix is specificity. Name the person you're making this for. Name the job they're trying to do. Name what would make them feel calm or capable when the product works. Name the last tool that let them down. If you can't name them by the win they're chasing and the worry they carry, you're making it for yourself, and the work will show it.

Most "user-centric" decks describe a persona. The good ones describe a person.

Good taste reads more than the user. It reads the various vectors influencing the market the audience sits in. What budgets are tight this year. What workflow this product is replacing. What promises the category has already made and broken. What the customer is hoping for and what they've stopped hoping for. Where the competitor is overinvesting and where they've gone quiet. Taste that ignores all this is technically clean and historically blind. It wins design awards. It loses customers.

An example. Two teams ship AI-powered meeting note tools in the same quarter. Team A nails the craft: beautiful summaries, clean export, tight integration. Team B's version is technically rougher, but it ships with one feature the others don't have: a "this meeting could have been a doc" detector that gently suggests when a meeting wasn't worth holding. Team A wins the design awards. Team B wins the market. The difference isn't the product. It's the room around it. A moment where every customer is exhausted by meetings and looking for permission to have fewer.

A product that was tasteful in 2024 can feel out of touch in 2026. The thing that changed wasn't the product. It was the room around it.

Before you can choose, you have to see. This is the layer everyone underestimates. It's slow attention. It's the catalog of references you've built up over time. It's holding multiple ways something could be done in your head at the same time. The categories let you see, and the more you look, the better the categories.

The 50-millisecond rule still surprises me: users form an aesthetic judgment about an interface in less than a tenth of a second, and that first judgment predicts how usable they'll find it more reliably than the actual usability does. That's how much perception runs everything else.

The disposition lives in the Compound: the willingness to be wrong out loud. This register is the craft moment that disposition produces. Choosing against other good options, under time pressure, without all the data. Both are required. A team can have leaders with deep conviction who freeze under deadline. A team can have leaders who decide quickly without the conviction to defend the decision a week later. The first ships nothing. The second ships things they abandon at the first criticism.

Most teams I've worked with drown when AI gives them more options than they had time to evaluate. The teams that ship are the ones who treat decisions as something to make and move on from, not as a chance to be perfect. The decision register is the one that gets harder as the option space gets larger. AI made the option space larger. So this register matters more now than it did three years ago.

This is where you learn the grain of the material you're working with. AI has its own grain: confident where it shouldn't be, vague where specificity matters most. Craft, in this era, means knowing where the model will fail you and putting guard rails there. It means writing prompts that force structure onto outputs that would otherwise wander. It means sweating the small thing that no one else will notice: the right word in the right place, the timing of the loading state, the choice between two acceptable defaults. That's where taste actually shows up in a product.

This is the layer where principles have to survive real tradeoffs. Patterns have to spread on their own without someone in the room to enforce them. Governance has to protect the work as it scales. Most projects I've been part of fail here. The principles on the wall die the first time clarity and ambition fight in the same meeting.

A real system is one where the principles show up in the actual product, not in the deck about the product. This layer doesn't photograph well. It's also the layer that decides whether anything else holds up.

"We prefer clarity over delight" is a real principle. "We love great design" is wall art.

This is the test: do the people on your team make decisions you would have made, when you weren't there to see them make it? Taste spreads through proximity, not through training documents. It spreads in the rooms where decisions get made out loud. Run your design crits as decision meetings, not feedback meetings. Run hackathons so the team gets a hundred reps in a week. Show off your team's work in your reviews, not yours.

You don't have a culture of taste until people on your team have started disagreeing with you, and being right about it. I'm still learning to enjoy that part.

For every reference you save, slow down enough to ask which specific part of it is doing the work. The whole landing page isn't the reference. The way it handles the third scroll is. The whole onboarding isn't the reference. The single screen where it explains pricing without a salesperson is. Find the part. Write down what it's doing and why it works. "This works because the load order surfaces the highest-value action first." If you can't write that sentence about a specific part, drop it. Mood boards are scrapbooks of things you find pretty. A reference library is a slowly assembled argument, one observed, named part at a time. This is the foundation everything else stacks on.

A reference you can't argue for is a reference that won't help you when you actually need it.

A hundred real decisions this month, not ten. The throw-away is where taste lives. Vibe-code a dozen variants of a feature. Throw away eleven. Treat the throwing-away as the actual workout, not the failure. AI made this kind of practice cheap for the first time in my career. Use it as resistance training. Don't use it as a shortcut.

The hard part isn't the making. It's the killing. Most of us, me included, on more projects than I'll admit, get attached to the work we just made and find a reason to keep it. This is the moment conviction actually has to fire. Look at the eleven variants. Be honest about which one is genuinely better. Kill the other ten without negotiating with yourself.

Without it, the crit becomes a place to talk about how things look. Design's version of group therapy, with everyone leaving lighter and nothing actually changing. With it, the crit becomes the place taste gets practiced in public. A feedback meeting produces comments. A decision meeting produces a direction. There's no other meeting on a design team that does this much work for the team's culture.

Try running one crit without the deciding question. Then run the next one with it. You'll feel the difference in the room.

Every design decision is answering a question on behalf of someone. A specific person, in a specific moment, trying to accomplish something specific, and quietly hoping it won't be one more thing that disappoints them. If you can't name that person, by the job they're doing, the win they want, what they're worried about, and what burned them last time, you're making it for yourself. Vague users produce vague work, every time.

"For users" is a dead phrase. "For Maria, who runs operations at a 200-person logistics company, trying to close out the quarter without surprises, hoping this rollout goes smoother than the last one" is a live one.

AI is impressive at making variants. It's bad at picking which one should ship. Use it to widen your option space, to iterate faster than you used to, to test ideas you'd never have had time to sketch by hand. Don't let it make the call for you. The work at the end of the pipeline is yours. Sit with the variants, choose what's worth keeping.

The team that uses AI to generate ten options for a human to choose between is doing real work. The team that uses AI to generate the right answer is fooling itself, usually beautifully and always confidently. I've watched myself do it. The model gives you a polished answer, you read it, you nod, you ship it. A week later you can't quite remember why that was the right call. That's the tell. If you can't reconstruct the reasoning, you didn't make the decision.

Three things get called "principles" and only one of them is. Take three sentences a real product team might write down:

1. "We love great design."
2. "We aim to ship products that delight customers and drive measurable business value."
3. "When delight and clarity conflict, we choose clarity."

The first is a value, a feeling about who we are. The second is an intention, a hope about where we're headed. Neither survives a hard meeting. The third is a principle, because it gives you the answer when the meeting actually happens.

Most company principles are auditioning to be liked. The real ones do work in the room when two good things fight.

Put your strongest people into rooms they haven't been in before: with executives, with customers, with the people doing the actual work the product is supposed to help. Put the people earlier in their careers into rooms where senior leaders are deciding out loud, with the messy reasoning still intact. Exposure is the mechanism. It's the slowest kind of learning, and the most durable. Stop trying to write the playbook. Start picking the room.

A culture of taste isn't built in a workshop. It's built by putting people where the decisions actually happen, often enough that the way the decisions get made becomes the way they think.