As more product work is generated from shared patterns, designers need to decide where an experience should surprise, celebrate, or simply leave someone alone.
Imagine asking an AI tool to design a task-completion state.
The result has a checkmark, a friendly message, and a small burst of confetti. It uses the right components. The spacing matches the system. The animation is polished.
Now imagine that same response after two different tasks.
In one, someone has finished setting up a business they have spent months trying to launch. In the other, an administrator has updated the forty-seventh record of the morning.
The component is consistent. The experiences should probably be different.
Nothing in the confetti explains which moment deserves it. That decision depends on what the person has done, how often they do it, and whether they want the product to make an occasion of it.
As AI becomes more capable of assembling interfaces from established patterns, this distinction deserves more attention. We are getting better at telling tools what belongs to our product. We also need to decide what belongs to the moment.
Different moments call for different levels of acknowledgment. Zillow’s success-state starter kit makes room for both quiet confirmation and celebration. Image credit: Kudzai Chinyadza, Celebratory Success States.Matching the system is an incomplete definition of quality
There are good reasons to care about consistency. People should not have to relearn familiar actions on every screen. Teams should not spend their time reconstructing decisions already made elsewhere.
Figma’s guidance on working with its agent describes how components, guidelines, examples, and task-specific context help generated work reflect a team’s product. That is useful progress. It also makes clear that consistency depends on the context supplied to the tool.
But a faithful rendering of a system can still be an emotionally inappropriate response.
A success message can be correct and patronizing. An illustration can match the brand and occupy space someone needs for their work. A playful loading state can be charming until it appears during a time-sensitive task.
These failures are difficult to identify by inspecting the component alone. They emerge from the relationship between the component and the person encountering it.
In my earlier article about design-system judgment, I explored the distinction between enforcing rules and deciding whether a pattern is appropriate. Expression brings another version of that problem into view: a detail can follow every rule we wrote and still misunderstand the occasion.
I want software to have room for color, humor, beauty, and small surprises.
I want an interaction to feel satisfying, even when satisfaction cannot be reduced to a shorter completion time.
I also want the product to recognize when I have work to do.
Pleasure has a history beyond the AI conversation
Designers were studying this long before generative interfaces.
Don Norman’s work on emotional design considers appearance, the experience of using something, and the meaning people attach to it. That gives us a broader vocabulary than “add delight.” Pleasure can come from how something looks, how well it responds, or what owning and using it means to someone.
Therese Fessenden’s distinction between surface and deep delight is useful here too. An amusing detail can create a moment of pleasure, while a well-considered workflow can produce a quieter satisfaction that lasts through the task.
The second does not make the first disposable.
A product can be dependable and exuberant. A useful tool can have a peculiar illustration or a lovely transition simply because someone cared about making it enjoyable. We do not have to defend every expressive choice by pretending it saves three seconds.
But we do have to understand what it asks of the person.
A color choice might make a screen more inviting without adding a step. A celebration might demand attention. A joke might require cultural familiarity. A transition might make a relationship easier to understand — or force someone to wait.
“Delight” is too generous a label to apply before finding out which of those things happened.
A playful product still needs restraint
Duolingo offers an instructive example because expression is already central to its identity.
In its account of redesigning the app’s core tabs, the team describes experimenting with several visual directions, then finding tensions between consistency and purpose. An illustrated header on every tab created uniformity, but some illustrations occupied space without communicating anything useful.
The solution included different header sizes for different purposes, with titles kept in consistent positions.
That is a more useful lesson than “make every screen playful.” A recognizable product can vary its emphasis while preserving familiar structure. The team’s own account shows that even an expressive brand needs to ask where an illustration earns its space.
Credit: DuolingoThis matters when a design becomes an example for AI to reuse.
If the reference screen has a large illustration, should the next screen have one too? Is the illustration a general rule, a possibility, or a response to a particular situation?
A tool can reproduce the visible answer while missing the reason for it. So can a designer working from an incomplete specification.
As generation becomes easier, that missing explanation can travel further. A choice made for one meaningful moment becomes the default for several ordinary ones.
The expressive detail survives. Its purpose does not.
Consistency can include a change in tone
Mailchimp’s voice-and-tone guidance makes an important distinction: its voice remains recognizable, while its tone responds to the reader’s circumstances.
The guide welcomes humor but gives clarity priority. It asks writers to consider whether someone has finished a campaign or is confused and looking for help, and warns against forcing a joke.
This is a practical way to preserve personality without requiring the same emotional performance everywhere.
Credit: MailchimpA product’s identity can live in how sensitively it changes.
Warmth during onboarding might mean encouragement. Warmth after an error might mean a direct explanation, preserved work, and a clear next step. The writing can sound like the same company without treating both situations as opportunities for comedy.
That distinction should also shape instructions given to AI.
“Make it friendly and playful” leaves a great deal unresolved. Friendly to whom? Playful at what point? What has just happened?
A more useful instruction describes the situation: the user’s work is safe; they need to understand why the action failed; avoid humor; make the recovery path obvious.
There is still room for good writing within those constraints. The constraints establish what the writing owes the person.
Expression can make an interface clearer
It is tempting to organize this debate as a negotiation between expressive design and usable design. Google’s research on Material 3 Expressive complicates that framing.
In its published research account, Google describes an email interface where the Send button became larger, more prominent, and closer to the keyboard. Participants located it faster. The expressive treatment also changed hierarchy and placement; its benefit cannot be attributed to decoration alone.
The same report includes a less successful experiment: a playlist replaced familiar rows with a scattered arrangement of album artwork. Participants found it visually exciting, but usability suffered.
These are findings from Google’s tested designs, not evidence that every product should adopt the same aesthetic. Together, they show why the actual interaction matters more than the adjective attached to it.
Credit: Google DesignColor can establish emphasis. Shape can distinguish actions. Motion can show where something went. An expressive decision may help people understand a product while making it more pleasurable.
The question is what remains legible after we introduce it.
A familiar control can carry considerable character. A surprising arrangement can remove the very cues someone needs to use it. Designers have to examine both effects, rather than accepting “on brand” or “more delightful” as sufficient explanations.
The same delight can become someone else’s interruption
Asana’s celebrations show why preference belongs in this discussion.
Its 2016 introduction of an expanded celebration system described occasional animated characters appearing after task completion, with celebrations enabled through settings.
A 2018 community discussion captures the other side: a customer found the characters intrusive and asked how to disable them. A reply explained the setting.
That exchange does not establish how most users felt. It does show that the same detail can be welcome to one person and unwelcome to another.
Credit: AsanaThis is where the word “delight” can hide an assumption. We describe our intention as though it were already the user’s response.
A celebration designed to recognize accomplishment may become an obstacle during repetitive work. Something enjoyed privately may feel inappropriate during a presentation. Someone may like the character and still prefer to finish without watching it.
Giving people control is part of designing the experience well.
For motion, the stakes can extend beyond preference. W3C’s guidance for Animation from Interactions, a Level AAA criterion, explains why people need ways to disable nonessential interaction-triggered movement, including the potential for vestibular symptoms.
An expressive experience should have a considered reduced-motion version. Removing movement should leave clear feedback, meaningful hierarchy, and an understandable result.
There are many ways to make something feel good. Movement is one of them.
AI changes how easily the wrong detail can spread
None of these problems began with AI. Teams have copied patterns without understanding them for years.
What AI introduces is another way to reproduce those decisions — and an opportunity to be more explicit about them.
A library might describe a celebration’s colors, duration, and animation curve. That describes the asset. It says little about whether a celebration should happen after someone completes a tutorial, pays a bill, or removes an account.
Those are different events even if the system labels all three “success.”
Here is a hypothetical rule worth testing: a fuller acknowledgment for a meaningful first achievement; brief feedback for repeated routine actions; direct, restrained language for sensitive outcomes.
That rule will not fit every product. Its value is that it gives the team something concrete to question.
Is the first achievement actually meaningful to the person, or merely valuable to our onboarding metrics? Does a repeated action become meaningful when it completes a larger goal? Are we assuming we know how someone feels because we know which button they pressed?
A good specification should leave those questions visible.
AI can help explore expressive alternatives: different copy, motion treatments, layouts, and levels of emphasis. There is no need to claim that surprise or beauty belongs exclusively to human production.
But producing an alternative and establishing when it is welcome are different kinds of work.
If we evaluate generated interfaces mainly by how closely they match the library, we will favor what is easiest to compare. If we ask whether the experience suits the occasion, we need examples, use over time, and feedback from the people living with it.
That is a broader evaluation than a screenshot review can provide.
“Worth caring about” should not mean harder to leave
There is another complication: products can use expression to serve goals people did not choose.
A cheerful character might encourage a useful habit. It might also make declining an offer feel like disappointing someone. A reward animation might acknowledge progress, or distract from how little progress actually occurred.
Those possibilities need to remain in the argument. Emotional design is capable of persuasion as well as pleasure.
The appropriate question is whose purpose the detail advances.
Does it help someone enjoy an activity they value? Does it acknowledge something they accomplished? Or does it make the product’s preferred behavior harder to refuse?
Time spent and repeat visits cannot answer that on their own. Someone may return because the product is useful, because leaving is costly, or because a mechanic creates pressure. Those explanations have different implications for design.
I am interested in software people appreciate using. That requires taking their ability to dislike a detail seriously, even when it is the detail the team loves most.
Give the expressive decision a proper review
In a product team, this work needs a place before the final polish pass.
For a proposed expressive moment, I would want to see it in three conditions: the first encounter, repeated use, and a situation where the person is rushed or something has gone wrong.
The same animation may behave beautifully in isolation and become tiring in a sequence. The same line of copy may be reassuring on a quiet afternoon and dismissive during a failure.
Research should make room for those differences. Watch whether people can continue. Ask what they think the response means. Find out whether they want to see it again. Include people who prefer less motion or sound, and test whether the experience still communicates clearly.
These observations will not produce a universal delight score. They can help us distinguish a detail people value from one they tolerate.
This also changes what we document.
Alongside the approved example, record the circumstance it serves, when to scale it back, and a case where it should not be used. Give both people and AI tools a reason to choose differently.
A design system can document when to hold back. Adobe Spectrum recommends restrained illustrations for errors, where an upbeat visual can clash with the user’s situation. Image credit: Adobe Spectrum, Illustration guidelines.A system with that kind of guidance can protect character. It can make room for an unusual illustration, a distinctive rhythm, or a moment of celebration while keeping everyday work dependable.
The alternative is to encode only what is easy to repeat and hope the rest survives delivery.
I want the products we build with AI to have more room for pleasure: a color choice that lifts a screen, a response that acknowledges effort, a transition that feels unusually right. I also want that pleasure to survive actual use, beyond the first demo and the designer’s own preferences.
Return to the two completed tasks. The new business owner and the administrator may both need confirmation. One might welcome a moment to take in what they have accomplished. The other might want the next record ready.
AI can generate both responses.
We still have to notice that the occasions are different.
AI can generate the interface. What makes it worth caring about? was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.