The Most Awkward Era for Designers Is Precisely the Most Important Era

Let's first admit an awkward fact.

Over the past year I've noticed a striking asymmetry: product managers can already use AI to turn ideas directly into HTML and a few context files, programmers have been liberated by coding agents to the point of seriously thinking about "what should I do next," and only designers are still stuck in Figma dragging pixels frame by frame. Image generation has indeed gotten faster, but "designing a product end to end" has not sped up to the same degree. Writing copy, pairing images, generating a decent-looking interface—AI can do all of that; ask AI to think through for you "whether this feature should be built, whether this path should exist, who this product is actually meant to serve," and it will only give you passage after passage of seemingly reasonable analysis, then wait for you to keep asking.

The sharpest part of this awkwardness is this: designers were originally the group that treated AI as a creative partner earliest, yet today they often can't find the link where they are amplified by AI. Engineers can pair with agents and write in a day the code that used to take a week; PMs can turn requirement documents directly into clickable pages; designers, meanwhile, seem to stand in the very center of the line, watching both sides take off while the mockup in their hands is still being repeatedly confirmed—"do we want this color scheme, how many pixels should this corner radius be."

This is my observation, not Ian Silber's exact words. But Ian Silber, Head of Design at OpenAI, said something in a similar direction on the Lenny's Podcast: this is the best time in history to be a designer. He also acknowledged that engineers' productivity has increased tenfold or even a hundredfold because of AI, while design teams' has not. The pain designers feel now is not that AI has taken their jobs, but that the role is becoming blurred—they don't know how much value is left in the few mockups they draw every day.

Design was never that mockup

Figma, Sketch, and code are all just media for expressing a vision. The medium changes; the judgment does not. Designers used to be accustomed to drawing everything out because drawing was then the only way to make an idea land and let others see it. But when AI can directly generate interfaces, you realize that "drawing it out" is itself the fastest thing to be replaced—because it is precisely the most mechanical, most replicable link in the whole chain.

Ian put it more concretely in an interview with Dive Club: designers will spend less time in pixel-pushing tools like Figma and more time thinking about how people interact with models—because the model itself is the core product. He gave a very telling example. For new user onboarding, the traditional approach is to design a set of onboarding UI that teaches users step by step how to use it. His team did the opposite: instead of adding an onboarding interface to the model, give the model enough context to know that the person on the other side is a new user who needs to be guided by the hand. Design did not disappear; it simply shifted from "drawing a tutorial page" to "deciding in what posture the model should welcome a person."

Once design is understood as "that UI mockup," AI indeed crushes humans. Once it is understood as "what this product is and how people and it get along," that mockup is merely an export at the end of the chain. For the same interface, the model generates ninety, and the remaining ten—what to include, what not to include, what the defaults are, what the first action guides users to do—is the judgment left by humans.

Why designers are actually more important

To say something potentially offensive: AI can already produce competent, even excellent, product design. What it excels at is recombining existing paradigms, assembling common interaction patterns quickly and presentably. You want a settings page, a data dashboard, a landing page—it gives you ten versions, each more standard than the last. What it lacks is not craft but judgment—observation of real people, a clear point of view, and the courage to invent a new interaction language.

Truly new interactions have no training data. The iPhone's multitouch, Snapchat's deliberately inverted communication model, Instagram's bet on mobile sharing—none of these were assembled from existing data. First someone had a judgment about "how people should use this thing," and then everything followed. Ian says that when anyone can make anything, the human side becomes the most important dimension of a product. Following this logic, what designers truly need to sell is not the deliverable, but that set of judgments that no one has yet validated but that is worth trying.

There is also pressure coming from below. Design used to be a specialty of a few; now PMs also have AI produce a first version and toss it to designers to "optimize." You'll find that AI's version isn't ugly—it may even be neatly laid out and logically coherent. This instead pushes designers into a more comfortable position: since AI has already made "competent" a standard configuration, "why is this worth doing" becomes the only thing designers can provide.

I call the experience an "invisible moat"—this is my extension, not his exact words. But his direction is consistent: when model capabilities are open to everyone, what distinguishes a product is no longer which model you call, but what you make of the experience. Models are shared; experience is personal. Others can copy your code, but they cannot copy your understanding of a scenario.

Great design often starts with refusal

Snapchat is the best example. Messages that vanish after viewing, opening straight to the camera, Stories replacing "select all and send." It is said that users frantically sent snaps to each other and clamored for a "send to everyone" button, but the team worried that would ruin Snapchat's tone, so they invented Stories—making content visible to everyone into narratives arranged by time that disappear that same day. This example is often discussed as a "social feature," but it is essentially a design judgment: the user's requested need is real, but the way to satisfy it is almost certainly wrong.

Instagram is the other side. Systrom cut a feature-bloated check-in app, Burbn, down to just three steps: take a photo, apply a filter, share. One detail is very moving: filters were born because his girlfriend felt that photos taken with the iPhone "weren't worth posting." What a product is ultimately remembered for is often not what it did, but what it restrained itself from doing. Burbn had so many features that no one could clearly say what it was for; after the cut, the whole world remembered what it was for.

These two cases are especially worth savoring in the AI era. Because generation brings the cost of "adding one more feature" close to zero, desire will be infinitely amplified: since AI can conveniently generate a check-in, a chat, a store, why not do it? But a product's texture depends precisely on how many of these temptations you resist. Great design often starts with refusing a convenient request—this statement is closer to truth in the AI era than before.

System design is more durable than one-off features

Notion made "everything is a block" its philosophy. Text, tables, boards, wikis are all composable blocks, and the slash command turns insertion into muscle memory. This system design is more durable than any one-off feature: a feature specifically for "project boards" may become obsolete in three to five years, but the primitive that "all content can be broken into blocks and freely reassembled" has supported Notion for over a decade. It also suits AI better—when a model needs to reason about a cohesive whole rather than jump between isolated features, composable primitives are naturally a better interface.

Cursor proved another thing: in AI products, differentiation still comes from interaction. Tab accepts suggestions, and continuing to type ignores them; Cmd+K edits code in natural language; the diff view lets you decide line by line whether to accept or reject. The same model capability is designed into a low-friction, reviewable collaborative experience. Everyone can buy the model; what you design "writing code together with humans" into is what belongs to the product itself.

Looking at these two together is interesting. Notion's durability comes from "structure first," while Cursor's value comes from "interaction design." One wins on system, the other on experience, but both point to the same conclusion: design is not icing on the cake after generation; it determines how far this tool can accompany you and whether it can hold up in real workflows.

If you can avoid designing it, don't design it

Ian's counterintuitive instruction to his team is "just do less"—if you can avoid designing it, don't design it; if you can reuse existing components, don't build new parallel features. At first this sounds like laziness, but it is actually the hardest lesson. Because "adding" is always easier to report than "subtracting," and making something new always feels more present than saying "this doesn't need to be built." Teams that truly build good products often spend most of their time arguing about "what should not be done."

He also mentioned a two-speed approach: durable interfaces deserve slowness—try a hundred, throw away ninety-nine, verify repeatedly; fast interfaces need four hours from idea to launch, because the underlying models change in weeks, and it isn't worth being overly attached to interactions that haven't yet proven durable. Behind this is a reverence for time—spend slow effort where slowness is deserved, and leave speed for the parts that will change anyway.

I often tell friends that true design is a kind of balance. Between what can be done and what should be done, between speed and durability, between expression and restraint. AI has pushed "what can be done" to nearly infinite, so "what should be done" becomes the only scarce judgment. In the past, designers' ability was shown in being able to make things; in the future, it will be shown in whether they dare to, and can clearly articulate, "this should not be done."

Back to the awkwardness at the beginning. Designers have not failed to be liberated by AI; rather, the old deliverable is losing value—that mockup, that set of pixels, is becoming cheaper and cheaper. What truly needs to be amplified was never output, but judgment: what is worth doing, what should stop, and who this product actually wants to stand with.

This is precisely the most awkward era for designers, and also the era that most needs judgment.