Claude Design shipped last week, and the entire internet had takes. Some hot, some cold, plenty from people who've never designed anything, all announcing that Figma is dead, or that design as a profession is dead. Anyway, tools evolve. That's the nature of technology.
I want to focus on what hasn't changed.
Two kinds of judgment still determine whether a product is any good, and neither of them has moved. The first is the judgment at the front of a project. The curation that decides what to build and, more importantly, what not to build. The discipline that keeps a product coherent when anyone with a model can generate anything. The second is the judgment at the end. The refinement that turns a competent first output into something that actually feels incredible. The discerning iteration that turns a rough draft into a perfectly honed object with emotional resonance.
Both are taste. And as the tools evolve, taste is what doesn't change.
Taste Isn't Easy
A small number of people are born with taste, an innate eye for what works. But most people who have it, needed to cultivate it. They cultivated it through years of obsessive attention to detail, a working memory of fashion, art, architecture, typography, and industrial design, and an emotional response to form they never quite outgrew. As Steve Jobs stated, "it comes down to trying to expose yourself to the best things that humans have done, and then try to bring those things in to what you're doing." Taste cultivation is slow, deliberate work. It requires effort, dedication, intentionality, and passion. If it doesn't matter to you, like deeply matter, it doesn't happen. Taste is also cultural, and the standards of taste are ever changing. This is why we have taste makers in fashion, graphic design, industrial design, and product design, really any design industry. Elevated taste is why product managers don't make for great designers. It's also why it's so hard to become an exceptional designer.
AI slop is what design looks like without it. The production is fast, the output is plausible, the judgment is absent. That's the gap the best designers live in. Some designers are in the profession because they like the process, without ever building the eye. The ones with a real obsession with taste use the tools as an accelerant. More experimentation, more exploration, more iterations toward something exceptional.
When the cost of producing design approaches zero, only quality matters. And quality is a taste problem. Maybe one day AI will one-shot experiences that land at the highest level of craft, with taste as a parameter you can dial. A confidence score for "beyond great." We aren't there. Until we are, taste is the thing AI can't fake, the thing that takes years to build, and the thing that matters more than ever.
The Two Sides of Taste
Taste does two kinds of work. Both get more valuable as production of design and code gets cheaper.
Taste creates. On the creator side, taste is what lets you reject the model's first output. It's the internal response that says not yet when everyone else says ship it. When anyone can generate a plausible layout in five seconds, the differentiator is the eye to see what's still off. Spacing that's wrong, tone that's cheap, motion that feels like a stock effect. Taste is the gap between competent and exceptional. AI moves the floor up. The ceiling still belongs to the people who can recognize the difference.
Taste curates. On the curator side, taste is what lets you decide what not to build. This matters more as production collapses. When a single designer with AI can generate a feature in a day, the temptation to ship everything grows. So does the organizational pressure. And most of what gets generated shouldn't exist.
Every feature you ship is a commitment. When the cost to ship collapses, the cost of bad commitments rises. Accumulation is how products decay. Vitamins crowd out painkillers. The app grows heavy with things that should never have been built.
Instagram prunes its feature set every quarter. Linear has stayed famously narrow despite shipping capacity that dwarfs its scope. Both are taste decisions. They come from a clear sense of the product and the willingness to reject anything that dilutes it.
In the new world, the discipline to say no, with conviction, with specificity, with evidence, is as valuable as the ability to create. Maybe more. The people who can only create will flood the market with forgettable work. The people who can create and curate will build the products that people actually love.
AI as an Amplifier
Before going further, a distinction is worth drawing. Craft in product design has five dimensions. Visual design, animation, voice, tactility, interactivity. Each has its own principles, its own history, its own way of being done well or badly. Taste is the thing that evaluates each dimension. It's the judgment that says whether the animation curve is right, whether the voice rings true, whether the spacing has rhythm, whether the interaction feels alive. Craft is the material. Taste is the eye that works it. And AI is the amplifier in the middle.
A designer with taste can now explore more of the craft space in an afternoon than a team could explore in a week two years ago. At Nubank, two designers used Figma Weave (which is a great tool I use for the illustrations in my articles) to create beautiful animations for four hero illustrations in half an hour. Previously this would have taken years of training in After Effects coupled with days of production. The iteration loop that taste demands gets cheaper and tighter. That's genuinely transformative.
Amplifiers amplify whatever signal you give them. The signal is the taste you apply when you evaluate the output and steer the next generation. A designer with taste rejects the first output and pushes the model toward something better. A designer without it accepts the median because it looks fine. Same models, different operators, different results.
This creates a paradox. AI makes it dramatically easier to reach "good enough." It also makes it easier to stop there. When a generated layout looks passable in five minutes, the temptation to ship it is real. The distance between "this works" and "this is loveable" stayed the same. The distance between "nothing" and "this works" collapsed. Teams that mistake the first milestone for the destination will produce mediocre software.
How to Build a Culture of Taste
Taste is the product of deliberate conditions. Talent is the raw material. The conditions around that talent shape it. And those conditions start before anyone touches a tool.
Start at the top. When care for craft starts with leadership, it becomes infinitely easier for teams to produce work with a high level of taste. When your CEO believes in craft, the friction around attaining it disappears. Stewart Butterfield wrote "We Don't Sell Saddles Here" weeks before Slack's preview launch, setting expectations from the top that taste and craft matter. Give your executives the language for both. Show them why it works. When they can see through your eyes, they can advocate in every prioritization conversation.
Train taste before tools. Teams need curatorial judgment before they get generative superpowers. Invest in exposing your team to excellence. Study what makes great software feel great. Study physical design too. Industrial design, architecture, typography, the art history that shaped the people who shaped our field. Build a shared vocabulary for why something succeeds or fails. Then hand them the AI tools. The sequence matters. A team that learns taste through the model will develop the model's taste, which is the median. A team that learns taste through the canon will develop their own.
Engineers carry taste. Designers and engineers have to build craft together, and engineers need taste at the same level designers do.
AI tooling is collapsing the functions on both sides. Designers can now build working prototypes without an engineer pairing with them. That's been a massive opportunity, letting designers explore ideas end-to-end at unprecedented speed. Product managers and engineers can now generate designs without a designer in the loop. Both sides of the old handoff are dissolving.
One thing has stayed the same. The final published code is owned by engineering. Some designers check in PRs. More will, as the tooling matures. Still, the systems we architect, the codebase we maintain, and the release we ship all belong to engineering. That means the last mile of taste lives with the engineers. The highest level of polish happens when the engineers building the product have taste at the same level as the designers.
The story of the iOS keyboard is the clearest case study. Ken Kocienda, the engineer who built it, created the first version and iterated with his team every day for a month. They tasted the soup together. Then a designer named Bas Ording came in for a day, provided some animation direction, and Ken went back to iterating. The taste of the engineer drove the experience. The designer provided spice, and the engineer was the chef. For more on how taste gets applied through iterative engineering judgment, read Kocienda's book Creative Selection.
It's also why the medium where taste gets applied should be code. A live, interactive, data-driven prototype that the team can play with and debate together. At Slack we had a product principle called "Prototype the Path." It was one of the most impactful principles at the company, and one the team practices to this day. You can learn more about this principle from a talk I gave on the topic at Config in 2023.
Curate ruthlessly. Apply a rigorous lens to what you actually ship. AI-generated craft is a starting point for refinement. Treat AI output the way a sculptor treats a rough block. It gets you to the shape faster. Dramatically faster. The shaping is still yours. AI handles the first 80 percent of the work. Taste lives in the last 20. Spend the hours you saved there.
Every quarter, look at what you've shipped. Ask two questions. First, what about this experience could only have come from us? If the answer is nothing, you have a taste problem. The product has been built by the median model everyone else is using, and you've outsourced your soul to it. Second, what shouldn't have shipped at all? Kill those vitamins. Protect the painkillers. The team's collective taste gets sharper every time they practice saying no to something plausible.
The Destination
In one of the most meaningful interviews I've seen on design, Jonathan Ive described working on a Sunday afternoon on what seemed like an absurd detail. The way a cable was managed inside a box. He knew millions of people would engage with that little tab. His reason for caring was simple. "I believe that when somebody unwrapped that box and took out that cable and they thought, 'somebody gave a shit about me.' I think that's a spiritual thing."
Steve Jobs called it expressing gratitude to the species. You make something with love and care for people you will never meet. And when they use what you've made, they feel that care.
AI is making competent software abundant. Every team with access to a model can produce something that works, looks reasonable, and passes a usability test. The floor has never been higher.
But loveable software is becoming rarer. Love comes from care. It's the last 20 percent. The intentional details. The craft that serves a purpose. The moments that make someone feel like the people who built this product actually thought about them.
What AI cannot do is care. It cannot decide that a moment of waiting deserves smooth jazz. It cannot feel that a checkbox should be the most satisfying interaction in the entire app. It cannot know that when I drag the pegman in google maps, his legs should swing right and left with physics that match the speed of the drag.
What it also cannot do is have taste. The taste to create something better than the default. The taste to curate out what doesn't belong. Those are still human jobs, and they're the jobs that determine whether a product gets loved or forgotten.
The organizations that treat AI as a craft accelerator, a way to explore faster, iterate cheaper, and spend more human attention on the decisions that matter, will build the products people love next. The ones that treat AI as a craft replacement will build products that are competent, forgettable, and indistinguishable from everything else.
People deserve an experience made with love and built by people with taste. AI just makes the choice starker.