Seven Theses on How the Designer's Profession Will Change

Introduction

A couple of days ago I ended up in a conversation with Alexey Yudin — a talented designer and a UX/UI teacher at UrFU. Alexey came to talk about his graduating students, and instead we had a spontaneous hour-long conversation about the fate of the profession and about how to teach it.

I dumped on him everything I think about this. And along the way I formulated it for myself.

In short: here are seven theses about how artificial intelligence is changing the designer’s profession.

1. The subject of design is the scenario, not the screens

I have always said that the subject of design is the scenario, not the screens. But before, it was easy enough not to notice that and to ignore it.

Something like: “well yeah, we do think about the scenario there, about the user, about their journey. But in the end we draw screens in Figma, right?”

So what came out of it was that no matter how much we said we were dealing with the scenario, the actual subject of the work was — screens.

And now everything has changed. Because working with an interface means directly describing the goal, the criterion, the user and their scenario in words.

And everything simply comes to the state it should have been in all along: designers deal with scenarios. Well, the ones who are still needed for something in a world where neural nets can make a UI from a description.

Yes, today there is a pile of tasks where a neural net loses to a decent designer. Graphic design (you can’t make a logo of a normal level with a grid alone). A skilled designer will assemble the final UI style better in Figma. On picking typefaces and the expressiveness of a grid, us leathery ones still have something to say too. But that’s all technique. All of it will be eaten. And sooner rather than later.

How this works in practice:

Here I am making a startup about verbal feedback skills (if you’re interested — knock, I’ll tell you about it). And I need an interface for the trainer. Before, I would have drawn a mockup in Figma, run a hallway test on a clickable prototype, brought in front-end devs. In a couple of weeks — the first version. And only then, on real data, would I have seen what needed redoing. And then — a cycle for the next two weeks.

Today I open, say, Lovable (it’s a vibe-coding platform) and describe: “Here’s a person. In such-and-such a context. They have such-and-such a goal. They land in such-and-such a situation. They need to choose how to give feedback. Here’s how the system should react to their actions. Here’s the list of requirements. Here are the criteria. Here’s the analysis methodology”. And I go through several iterations of working with the text and the prototype. By evening I have the 5th or 7th version of an interface that has already been tested several times. On real data. It’s a perfectly workable program, capable of delivering value. It’s just that its code isn’t of a quality you could scale. But you can test it on live users, and see far faster and more vividly whether the interface helps the client solve their tasks. Not “imagine whether it helps”, but watch how they solve a real problem.

I think that today this is exactly what a designer’s work is. And screens are just a visualization of the scenario. And their 1st, 2nd, 15th version can perfectly well be vibe-coded. Yes, of course, afterwards all of it will have to be wrapped in an impressive UI. Once the UX has appeared as a result of evolution, in a collision with reality.

Why is a designer needed in this process?

If all they could do was make a beautiful UI — well, so that once PMF has been found, they can finally make a beautiful UI. But actually, it’s so that all these iterations can be turned, and the best interface solutions found and tested.

The principle: a designer’s work is creating and testing scenarios. Now describing the scenario in words is the method of work itself. Hooray!

2. The career trajectory has flipped over

In the first thesis I said that the principle stayed the same, it just became more important. But the career trajectory has changed substantially.

The sequence used to be like this:

1. Junior — you learn to work with your hands. Figma, banners, landing pages.
2. Middle — you learn to apply ready-made recipes. And to apply them in a specific context, for particular users, for given scenarios.
3. Senior — you learn to determine for yourself what is actually needed, you turn problems into tasks, you choose the stack the task gets solved on. And you take on more complex tasks, you work with large systems.
4. Art director — you learn to explain all of this to other people. For that you have to learn to detach the method from yourself. To describe how you do what you do. To say out loud with your mouth what you used to do with your hands. In such a way that people understand you. To formulate the quality criterion. To accept the work. In general, if a senior can explain to others what they do — that turns them into an art director.

Now you’ll have to start with the ability to describe things in words.

• If you can’t describe what you’re doing.
• If you can’t choose and describe the methodology you work by; if you can’t put quality criteria into words — you can’t work with the scenario. And that means there’s no place for you in the process.
• This level of abstraction used to belong to team leads, and here it’s needed by juniors.

And that, damn it, is hard. Because it’s a high-order skill. But without it you simply don’t control the process. How do you rebuild education so that juniors learn what is hard to teach a lead? A good question, damn it!

3. Generalists got a second wind

The use of AI in the work means a lot for the old argument of “specialists versus generalists”. All my life I have bet on super-generalists. I’m a super-generalist myself. I can run a business, manage a product, make interfaces, write texts, develop gameplay. And other things too.

But some time ago it started to weigh on me that I hadn’t chosen one single thing — so as to be the best at exactly that. Over the last few years I had been gradually moving away from the idea that generalism is the best of paths.

Artificial intelligence turned it all back.

A concrete example:

I needed to make a visual identity for AmpCamp. (This is me designing on the quiet for friends, for free — since the director isn’t allowed to design.) What would I have done three years ago? I’d have made a logo, a couple of forms, some blob for a T-shirt — and all in one version. And if I’d wanted something on a bigger scale, I’d have needed a crew: illustrators, animators, calligraphers, 3D people.

What did I do now? Four evenings of work. What I got was a whole spread of everything: video intros, renders of bandanas, merch graphics, animation for a video installation, mockups on socks and T-shirts. 38 variants of the pins alone. A crowd of variants for every single thing.

Did I need a 3D specialist in this story? An animation one? A graphic design one?

No. A calligrapher, now — that I did need: with calligraphy the neural nets are still hopeless. But to steer all of this, I needed to understand what video is, what graphic design is, what communication is. How to take the task off people, how to accept it, what method to act by.

In general — the market needs art directors, the market no longer needs just designers. It’s no longer enough to be able to do something with your hands, because now an art director can afford to do it all single-handed. He is responsible for the concept, for the integration, for making sure all the chosen artistic means match the communication task.

He did that before too. Nothing has gone anywhere. It’s just that before, he had to convey these thoughts to people in words, and now that is comparable in level of effort to making all of it himself. But for that you have to understand each of the operations somehow.

I don’t want to say there won’t be successful narrow specialists. No. Look, for instance, I order the pictures for my blog from Olesya. Not because a neural net can’t do that style (it can), but because Olesya has an awesome sense of metaphor. What I mean is: she doesn’t just draw brilliantly, she understands what she’s doing and why. She can put it into words.

4. What matters most is the number of iterations

The main thing AI changes is the number of iterations we can afford. How many variants we can look through. How many times we can reject what isn’t good enough.

This isn’t “you two out of the casket, do it for me, I’m too lazy to think what, decide for yourselves”. I mean — it’s not a reduction of cognitive load, not less work, but the possibility of creating a result of a different quality.

How this used to work:

The team busted its ass for a couple of team-months and put together one variant. Brings it to the client — the hypotheses don’t come true. If the assembled people have the time, the budget and the will — they’ll do one more loop. Fine, two. But not five loops. That’s way too much work, way too much time. In most cases everyone will simply give up. Purely from exhaustion.

How it works now:

Here we have a startup. I model the interface, I describe the idea of the scenario in words. Yesterday I went through five loops. Five iterations.

Made it → tested it on real data → saw it wasn’t right → changed it → tested it again.

And that fundamentally changes the quality of the result. Because you have more variants, stricter filtering, you can afford more.

An example:

We’re making a logo. The logo we make by hand — a neural net can’t handle the subtleties of graphic design, and besides, it doesn’t take long. Next I need to see how it works in different contexts.

Before, I simply wouldn’t have shown the client most of the applications. Because I wouldn’t have made them. Too expensive in time.

And now: I take ControlNet → feed it the silhouette, say “don’t touch the silhouette, add contextual elements around it” → get 20 background variants in 10 minutes → come up with another 20 variants of the logo on a medium → understand how it works → tune it → pick the best.

Here’s the thing: without these visualizations the team simply wouldn’t have understood the idea. We simply wouldn’t have gone down the chosen trajectory — it would have been too expensive to make all these variants just to check the idea. People aren’t good at perceiving abstractions, they order design in order to synchronize their mental pictures. Now this can be done in a wider context — you can afford more use cases.

5. Measurable quality versus imitation

And this brings us right up to the conclusion: the main thing to pump up in an AI world is the ability to choose a method, to set a task and to accept it.

In other words — to define and measure the necessary quality.

If you can’t measure quality — you don’t control what you get.

Your result resembles design. It quacks like design. But it isn’t design, it’s an imitation.

What the difference is — with slide design as the example:

It doesn’t matter how “beautiful” the slide’s styling is, or whether it hits the current fashion in interfaces. What matters is which thought it helps to understand or to feel, and by what means it does that.

And if there’s no thought behind the slide, if no emotional task is being solved, if we have no correspondence between the artifact and the task, because the criterion hasn’t been formulated and the result hasn’t been checked — it’s an imitation. It won’t help communication.

And at every stage you have to have a criterion: what’s good, what’s bad, why. If there’s no criterion — you’re just choosing what’s “prettier” or what “you like”. And that’s an imitation of design, not design.

Erudition versus competence

It has become very visible how big the difference is between erudition, command of technical techniques, and competence — that is, the ability to solve a specific problem in its context by available means.

Before, we could avoid noticing this. Because the designer worked with their hands, and it seemed that the main thing was command of the tool. You know how to do auto layouts in Figma? Good job, you’re competent.

And now it turns out that this was erudition. Technical command of the tool.

Whereas competence is when you understand the task, know the methods for solving it, can choose a suitable method and apply it. You can evaluate the result against criteria that match the task.

Here I’ll quote a piece of the conversation with Alexey:

Alexey:

— But I see the difference in quality between what a neural net puts out and what an expert can do. How subtly an expert can handle the management of human attention, using cognitive-science patterns, Z-shaped and F-shaped eye movement patterns.

Me:

— But nothing stops you from telling an LLM: “use the F-pattern”! The point isn’t that you know how to apply the method. The point is precisely that you can identify these patterns and understand when applying one is effective. And if that’s the case, then you can check what the neural net generated against those criteria. Fix it if needed. Prompt it correctly, if you know how.

It’s a completely different matter if you were simply handed a prompt that says, among other things, “apply the f-pattern”. Will that improve the result? We don’t know. Maybe. Maybe not. Here’s what AI made harder: now, in order to notice low quality, you need higher qualifications. It’s not enough to look at the list of applied methods — applying any number of them has become possible and cheap.

That is exactly the difference between erudition (I know what an F-pattern is) and competence (I see its absence in the mockup and I know how to fix that). And we’re going to have to change curricula so that competence actually does get formed.

6. How to teach in the new reality

As for teaching design — the fundamentals will have to be taught even more thoroughly.

It’s even more important to understand well, and to be able to explain, what the composition of a frame is. How it’s connected to the message and to the scenario. What that scenario is. Who the person is that we’re developing it for. And what goal that person has, and how it relates to the goal of the system’s existence.

And also to be able to learn — specific technical tools go out of date within a month.

What it’s too late to teach:

Don’t teach the previous approach. I mean — don’t start with Figma.

Here’s an analogy for you: when I was studying at the architecture school (the year ‘96), we were forbidden to use computers. Everything had to be sprayed with an airbrush, because that was true. And in ‘97 I became one of the first web designers in Yekaterinburg. For the first two years they taught us composition — thank you, that’s useful. A feel for the sheet, how to redraw a form, how to model a form — all useful. But what the hell are you teaching us on the stack from the era before last?

What is worth teaching:

Describing the task in words. Formulating criteria. Choosing a methodology and evaluating its applicability. Understanding user psychology. Modelling scenarios. Testing. Evaluating the quality of composition. Honing a sense of style.

And yes — professional erudition needs pumping up too. There’s an illusion that each of us now has borrowed erudition. We’ve got the best prompts, after all, the ones we downloaded from the vaults of the best prompts! But it doesn’t work that way. More precisely — it works very similarly, and that makes it harder to understand that the result is empty and inapplicable.

What matters is that principles can now be taught in practice. Today you can throw students straight into very varied practice and give feedback on that practice in the terms of theory. Because understanding something and making something have become very similar. So teaching means forcing people to analyse their own practice. Teaching them to notice why something works or doesn’t work, which principles have been violated, which methods have been used.

What’s next

I haven’t studied as much over the last 30 years put together as I have over the last two. All I do is learn. Here I am, off to yet another course — and that’s while teaching all of this myself.

At the end of our conversation we agreed to discuss further what can be done with diploma projects where AI and design meet. As you’ll have noticed, I myself don’t yet fully understand what follows from this for education. But the old model is out of date and a new one has to be invented somehow.

There’s a crowd of possibilities. But all of it has to be reinvented from scratch.

And, as I said above — there’s a good question: how the hell do you teach juniors what even leads can’t manage to learn in years?