Taste in the Age of AI

AI can generate polished interfaces, brands, and visuals faster than ever. But it still struggles with one of the things that separates work that merely looks good from work that actually lands: taste.

Before getting into why AI has no taste and where AI might be going instead, it’s worth defining what taste actually is.

Taste is good judgment under context

Good taste sits somewhere between foreign and familiar: foreign enough to feel interesting, but familiar enough to feel acceptable.

Push too far toward the foreign and you get discomfort, disgust, or something that reads as avant-garde. Fashion is a great example. A look can be genuinely new, technically impressive, and even ahead of its time, but still not land because most people aren’t ready for it yet. But if you push too far toward the familiar, you'll get boredom. It feels safe, predictable, generic. You’ve seen it too many times already.

Taste is the ability to teeter on the line between foreign and familiar intentionally.

Unfortunately, that line is never fixed. It depends on culture, timing, audience, and repetition. What feels fresh in one city, community, or moment might feel played out somewhere else. What looks strange today might become normal next year. Something can enter good taste at a specific moment, then exit it once it becomes too familiar. That’s why taste is hard to turn into a set of permanent rules: Its value comes from its relationship to the current cultural moment.

Good taste is effort made visible

That judgment doesn’t appear at the end of the process by accident. It comes from looking at references, exploring alternatives, and going far enough into the details to know what belongs. Effort is how you see it show up. The more effort you put into a product or brand, the more likely you are to arrive at a better answer, not because more time automatically makes the work better, but because time and effort creates room to explore. You reject the first plausible answer instead of mistaking it for the right one.

You can feel when someone has pushed on the work. They know when to make something stranger, when to pull it back, and when to leave it alone. They make deliberate decisions about what they choose to push, what they pull back, what they remove, and what they refuse to accept as good enough. Good taste is effort made visible. It’s selection, restraint, refinement, and point of view.

Bad taste is a lack of effort. Slop isn’t necessarily broken or ugly. It’s visibly under-considered: the work stopped at the first plausible answer. That’s why slop is so recognizable. It’s the visible absence of effort, and AI makes that absence much easier to see.

AI raises the baseline

AI also changes how effort shows up. People can get much further in less time, so the result can look more considered than pre-AI work even when the process involved less deliberate exploration. The default got better along with the speed. AI is so good at reading and reproducing old patterns that its standard outputs are now better than what most non-experts could create on their own.

But if the baseline is already polished, effort can’t mean producing something that merely looks finished. Effort is effort, after all. Most prompts don't take any effort. This means you have to be willing to push beyond what AI produces by default: exploring further, going deeper into the details, and taking the design past the standard answer even when that answer is already good. The standard output may be convincing without being tasteful. The effort is in using the time AI gives back for that exploration, not in accepting the first polished result.

AI can get a team to a competent baseline faster, but that only helps if the extra time goes back into exploration. If the time saved simply becomes a shorter deadline, the baseline improves while the work stays generic.

AI is also changing the creative timeline. Execution is becoming a blip. A first interface, identity, or set of visual directions can be produced almost immediately. That shouldn’t make the rest of the process shorter. Exploration, ideation, brainstorming, references, and critique now need to take up the majority of the process. Otherwise AI simply makes the generic answer arrive faster.

The mismatch

Models are already very good at generating familiar design patterns. Many UX problems have strong precedent now: onboarding, navigation, settings, forms, dashboards, checkout flows. AI has seen enough examples to generate something usable very quickly. But a polished interface can still arrive without taste.

AI learns from what already exists. It can reproduce patterns, combine references, and make variations at a speed no human can match. Taste often requires knowing when something has already become too familiar, when a visual language is exhausted, or when a new direction is just beginning to become culturally legible.

That’s the mismatch.

AI is fundamentally looking backward at what has been made. Taste is often about understanding what’s happening right now, and sometimes sensing what’s about to happen next. That’s why AI-generated design can drift toward the average. It can look finished and offend no one, and still have no real point of view. It can stack familiar patterns and features together without understanding the larger product, the cultural moment, or why one decision should exist over another.

The person directing it

AI is already useful for solving familiar problems, accelerating prototypes, generating variations, and reducing the cost of execution. With strong context, good references, and a research-first process, it can be genuinely helpful. The quality of the result still depends heavily on the person directing it.

The output improves when you spend more time in planning, gathering references, exploring directions, and giving the system enough context to understand the actual problem. Generate a screen and ship it, and you get to the baseline faster. Use AI as part of a design process, and you have a chance of getting past it.

I don’t think the goal is to inject taste into a model. Taste is too contextual, too cultural, and too dependent on the moment for that. Training on more historical examples or collecting more preference data may make models better at producing work people generally like. That’s different from knowing what’s timely, what’s stale, or what’s about to land.

What is entering the moment

If taste is partly the ability to read the moment, AI might be more useful helping us see what’s entering that moment than trying to have taste itself. Social media already has versions of this. Creators don’t just look at what’s popular. They try to spot what’s beginning to take off before it becomes saturated. The valuable signal is what might become culturally relevant next, not what everyone is already copying.

Design will likely move in that direction too. AI may get better at surfacing emerging visual languages, interaction patterns, audience preferences, and cultural signals. It may help designers get closer to the edge, where something is new enough to matter but familiar enough to land. Someone still has to decide what’s worth pushing, what needs to be pulled back, what belongs in the product, and what should be removed entirely.