A Conversation on
Design in the
Nascent Era
of Intelligent
Products.

Most of what gets written about AI and design falls into one of two camps. Either design is dead and designers should retrain, or design has never mattered more and AI is going to be great for everyone. Both positions are easy to hold and hard to act on.

On a Friday evening in February, thirteen designers and creative directors, plus a handful of product leaders, sat around a long table at Foreign Cinema in San Francisco for the first edition of Frontier Dialogues, an invitation-only dinner we organized to work out what is happening to design, and how the people doing it should deal with a technology landscape that won’t sit still.

The guests came from Bumble, Google, Meta, Monarch Money, Cohere, and others. Some had been designing for the web since before design systems existed. Others are building AI-native products right now. There were no slides or panels, and nobody had come with talking points.

Over several hours and a few glasses of wine, the conversation landed somewhere more interesting than either of those extremes. The honest answer lives in the messier ground between them.

People disagreed with each other and changed their minds mid-sentence. They would get excited about an idea and then undercut their own excitement a moment later. It was a good thing to be in the room for, a table of people who care about this working it out together.

What came out of it was a set of tensions. Real ones, the kind that don’t resolve neatly but that anyone making products right now will recognize. We wrote this piece to share that conversation with the rest of the industry, and to keep those tensions intact rather than smooth them over. Each chapter takes one. Read them in order, or jump to whichever one pulls at you hardest. Let’s get into it.

The evening opened on first principles. Part of why we wanted to have this dinner in the first place was a frustration with how the profession had calcified. The people who figured out web design by doing it, instinctively and vocationally, never stopped evolving. But the things they wrote down along the way got treated as scripture by the generation that followed. Boot camps turned process into dogma, and design slid from a living form of critical thinking into a set of steps you follow to produce a thing.

AI is blowing that up. The real question is what was underneath the dogma to begin with.

“What a designer does, if I go first principles,” Pablo Marques said, “is unlock computing power. A computer is capable of doing many different things, but you need someone that can instruct that computer to do that.”

The whole apparatus of product design, the wireframes and flows and systems and states, exists because users couldn’t simply tell a computer what they wanted. The designer stood in the middle and translated.

That middle is now compressing. “Computers are starting to be able to understand intent,” Pablo continued. “You can communicate in a more human way. So you don’t need programming or anything like that to get some of that computing power.”

He traced the idea to a talk he’d seen Stephen Wolfram give around 2013, where Wolfram described his ambition for Mathematica: letting people program a computer just by speaking to it. “I think in a way, LLMs are doing what he wanted to do. To a certain extent. But then how does that change what we do?”

Nobody treated this as abstract. Several people at the table were already working in places where the interface itself, the screens and the menus, was being bypassed outright. Conner Drew put the endpoint plainly: “Siri, add toilet paper to the shopping list. I get a delivery once a week. Zero UI. Zero anything. That’s where we’re heading.”

Nobody mourned it. The translation layer that designers built and maintained for decades is collapsing, and in some places disappearing altogether.

James Song thought the frame was too small. Where a line like Drew’s usually invites gloom, he heard the opposite. “The last major paradigm shift was the mobile phone, the form factor, and we’ve been innovating on it for almost 20 years,” he said. “Now with AI we have a new enablement technology, and if you lock your POV to just the phone and the web, of course it’s going to feel smaller. But glasses, world models, robotics are all enabled by this new technology too, and I’m hopeful that the dream of ambient computing, that everything in your environment will be tech enabled, will remove emphasis on a single device and allow us to reengage with the physical world armed with AI-enabled software.” His bet is that this revives whole disciplines: “the re-rise of things like service design, experience design and orchestration, instead of just competing with an LLM to design a better button.”

That sharpens the question the rest of the evening kept circling. If the translation layer is going, the job it used to define has to turn into something else. Into what?

If Chapter 1 was about what’s disappearing, this one is about what gets lost on the way.

John Roescher offered a metaphor that kept resurfacing all night. “When the user is able to put their effort into creating the outcome and experiencing the outcome versus the steps to get there, something changes,” he said. “I think of it as the WYSIWYG for everything.”

His example was the map. Planning a route on paper used to take real cognitive work. You had to read the terrain and keep all the turns in your head. Google Maps collapsed that into a single input. Most people now choose zero engagement with the act of navigating. They just go.

Roescher ran the same logic through tools like Adobe Dreamweaver, which once let designers flip between a visual canvas and raw HTML. “You could choose how much effort you put into the code side. But nowadays, take that out of the website and into any way you interact with computing.”

His point was that the craft has to migrate. If AI handles the process, what’s left to design is what the user meets at the end of it. “As designers, there’s an opportunity to put more effort into designing the outcome interactions than the steps to get there.”

Most of traditional product design has lived in the process, in the flows and states and transitions and error handling. If a prompt or a delegation swallows all of that, the interesting problems move to the other side of the interaction.

This is where some people got uncomfortable, and for good reason. If you skip the steps, do you lose the understanding?

Conner Drew described shipping a feature to production in Cursor, something he’d never done before. “I had something in my head. It allowed me to execute and remove the pretend layer and ship into production,” he said. “It’s empowering. It’s a very different skill set.”

That pretend layer, the prototype or the mock-up, has always done double duty. It communicates an idea, and it also forces you to think one through. Building a prototype slowly is how a designer finds out what they mean. Take the step away and you go faster, but you risk stripping out the slow reasoning that produces the best work. Drew put it in the body: “Why do therapists encourage you to journal by hand versus typing? Because it slows you down. It makes you actually aware of your thoughts and surface deeper insights.”

Speed and understanding are not the same thing, though for most of design’s history they came together. Now they come apart. The advantage goes to whoever can tell when a problem wants speed and when it wants to be worked slowly by hand, instead of running everything at one setting.

Peter Markatos made the night’s sharpest prediction. “You’ll be able to make a product just from a prompt. We’re probably a year or two away. And they’ll look pretty good. I think it’ll be a sea of similarity.”

He compared it to the jump from skeuomorphism to flat design, when every interface suddenly looked like every other one. “All the websites looked pretty much the same. Animated. Not in a good way, by the way. I think that was a pretty shitty period.” He expects the same flattening again, faster and at a larger scale.

He was also optimistic about what comes after it. “What that enables us to do as humans, as creatives, is the weird stuff. The design that’s off-kilter, superhuman, and super unique. It’s going to be the thing that’s differentiated. I’m actually betting on a creative renaissance that will come on top of all of this.”

Kevin Funkhouser wasn’t convinced. “My fear is that because it’s business-savvy, cheaper, faster, it normalizes,” he said. “Whenever you’re opening your phone, and literally all of the experiences look the same way, that is your definition now of normal quality. And what is the reason you’re going to be able to justify for craft differentiation?”

When “good enough” is cheap, the baseline moves. People stop noticing the difference because they have never met anything better, and once they can’t perceive it, good luck putting it in a budget.

Saurabh Kakade pushed back on that fear, starting by being blunt about what the work is. “Let’s be honest about what we do. Graphic design, creative strategy, marketing, it’s applied arts. There’s a brief, and there’s a commercial outcome expected from the work. That’s exactly what automation will keep getting better at. It’s always better than we think it can be; the rest is just a matter of time.” But where Kevin saw normalization, Saurabh saw a line being drawn. “When automation covers the commercial baseline, and that work can be genuinely good, a very clear line appears around where the human adds value. Work that comes from your own thinking, with no commercial angle: that’s art. And it gets valued more than ever, precisely because the separation finally becomes visible. You stop fighting to convince a client that the human part is worth it. The work separates itself.” The task, for creatives and for the companies built on applied arts, is to keep that artistic thinking flowing into the commercial work, because that is where the human value will sit.

Markatos and Maria Springer both worried about this generationally, from different directions. Markatos: “My son, who’s 15, he’s growing up with this. The skills we learned to make things in the past, those don’t matter as much anymore.” Springer’s worry was exposure: “A lot of their exposure is content, videos, whatever. If you’ve grown up with that, your definition of quality drops significantly.”

Alex Prodaniuk put the economic question directly. “Does the human touch, the paintbrush stroke, the imperfection, does that become couture and largely inaccessible? Or does it become just high-end?” Markatos’s answer was honest: “Depends on how good people are.”

Springer drew the structural line bluntly: “Amazon doesn’t give a shit about craft. They care about optimization. They’re going to run all the A/B tests automatically. But other companies are going to care.” Someone countered: “Amazon is pure utility. Apple is a luxury brand. They’re different businesses.”

Conner Drew added a data point from a product sense course he’d just taken. The instructor had asked which brands come to mind when you think of taste. “The only company listed that’s in the top 10 most valuable companies in the world was Apple. The rest was Amazon, Microsoft, Nvidia. None of them have taste. But that’s 90% of the top 10.”

James Song wasn’t ready to grant that cheap and normalized wins by default. “The truth of capital ‘D’ Design still holds true,” he said. “Make something that clearly matters to people, and the business will grow. If a business doesn’t want to take the time to understand their users and audience and just optimizes for imperfect metrics, you’ll end up like all the other mobile game developers that use dumb, annoying and artificial ways to juice their numbers. And the people will notice and reject them, that I believe.” For him the question that mattered was simple. “Anyone can make an app now, great, who is the BEST at it?”

So the market splits. On one side, companies that treat design as optimization and will gladly hand it to AI. On the other, companies where taste and craft are the product, where what makes people care can’t be generated on demand. The gap between those two worlds is going to widen, and the people most exposed are the ones in the middle, doing competent work that a machine can now do just as competently. Knowing which side you’re on, and being able to say out loud why craft matters to the companies that need it, is most of the job.

Underneath the bigger debates ran a quieter thread about behavior. Pablo Marques had put it on the table as one of the evening’s provocations: “A lot of interfaces are social contracts. If you think of the interface of Twitter or Instagram, someone’s going to post a thing, somebody else is going to see that thing, somebody else is going to like that thing. There’s a standard of interaction that creates a standard of behavior that then creates a whole social scene around that piece of software. Based on how the designer chose to design that.”

A like button. A feed that ranks one post above another. Neither is a neutral feature. They set norms. They shape how people perform for each other and what gets counted as taking part. A designer put those mechanics in a particular order with particular affordances, and millions of people arranged their social lives around the result.

So what happens when the interface stops holding still? Pablo pressed it: “If that gets out of the hand of someone who’s actually crafting it, would behaviors like that emerge naturally? Would you be able to create the same behavioral coherence when building products with generative interfaces?”

If interfaces generate themselves in response to context, you can get behavioral norms that nobody planned and nobody is answerable for.

Peter Markatos took this somewhere most public conversations about AI and design don’t. “We’re not making endpoints as much as we used to. We’re actually creating systems, and those systems, because they’re non-deterministic, become a thing where you, as a designer, as a maker, you have to reprogram and put guardrails around the morality of the system.”

He kept going: “That’s a totally new thing. You never really had to think about morality. Not as much when you design an interface. But now when you’re designing an agent experience, non-deterministic and totally open-ended and could wreak serious havoc on someone’s life, you have to be programming morality.”

Design has always carried ethics inside it. A confirmation dialog before you delete something. A warning before you message everyone at once. Those are moral choices expressed as friction. In a non-deterministic system, the same choices get baked into constraints the system applies on its own, with no designer reading each output before it ships.

Markatos reached for an unlikely reference to make the stakes land: Tucker Carlson. “I’m not at all a fan of Tucker Carlson,” he said, “but his interview with Sam Altman was very interesting. Most tech interviews of Sam don’t get into the morality of ChatGPT. Tucker’s very religious, and he dug into that. He straight up asked: who’s making the moral decisions at ChatGPT?” People laughed. The point underneath it wasn’t a joke.

Someone is already making these calls inside every company building AI products, and right now it is mostly engineers and policy teams. The argument from the table was that designers belong in that room too. Designers are not more moral than engineers. They are the people whose whole job has been how a person experiences a system, and that is the exact question an autonomous system forces into the open. When the system is unpredictable and runs on its own, how it ought to behave is a design problem, whether or not anyone is treating it as one yet.

The word that kept surfacing, unprompted and from different corners of the table, was soul.

“No one’s talking about the soul in these conversations,” Conner Drew said, reaching for the difference between a Gothic cathedral and a modern skyscraper. “Look at writing versus speaking. Why do therapists encourage you to journal by hand versus typing? Because it slows you down. It makes you actually aware of your thoughts. And I feel like it’s the same thing as designing something as elaborate as Notre-Dame versus a building today. The Empire State Building, still a gorgeous building, impressive. But it doesn’t have that same soul.”

Alex Prodaniuk grounded it in something physical. “There are inherent human qualities and attractions that are undeniable,” she said. “Sitting around a campfire. Hand-lasted footwear. It’s why mid-century modern furniture is so expensive. Forever and ever and always cool. Because there is something about a human spirit that is responding to something designed by a human spirit.” She paused. “That didn’t go away when we went from spending 300 years to build a Notre-Dame to putting up a building in a year. It did change the percentage of things that are considered to that degree. But it didn’t go away.”

Neither of them was being nostalgic. Once AI can generate competent work cheaply and without end, the things it can’t generate become the scarce inputs, the work that needed a person to mean it.

Conner Drew brought up something that had gone around that same day. Vogue Japan had run a cover of Ariana Grande, entirely AI-generated, without the artist’s knowledge, and the backlash was instant. “The inherent reaction, the negative reaction to something like that, is because of a lack of soul,” he said. “There’s a difference between a clearly AI-generated thing and pretending this is a human that you know and that you’re a fan of. Part of the reason you’re a fan of them is because of the energy, what they represent, and what they’ve done. And when you take that away, it’s fake. It’s a lie.”

Prodaniuk saw the same arc in social media. “Instagram started with beautiful pictures of your friends, how warm and fuzzy and cool and artsy it was. And then it became this machine for advertising. And now it’s a scary thing we’re all addicted to and also try to avoid.”

Rob Young pointed to the counter-movement already underway. “There’s this movement around personal messages, DMs. The raw photo. People are getting fed up with perfection. AI can produce it. What’s the difference? You’re hungry for that authenticity again. For that soul.”

Wanting soul and paying for soul are not the same thing. Prodaniuk named the trap: “A lot of that comes from slowing down. But how do you get businesses to want to invest in slowing down, and prove an ROI on that?”

Peter Markatos offered a historical frame. “When the photograph was invented, everyone thought it was the end of art. What it actually did was democratize portraiture to the masses. And then it freed painters to move into surrealism, expressionism, and all these abstract movements. Because photography captured reality, and it gave license to artists to have fun and play.”

Then he caught himself. “It’s a little obviously different now, because it’s machines that are becoming the photographers. But something like that.”

The analogy isn’t clean, but it holds up. People can feel when a thing was made on purpose, and they always could. What’s new is how wide the gap is getting between work somebody meant and work a machine produced without meaning anything by it. The hunger for the first kind isn’t going anywhere. The easier everything else gets to manufacture, the more people seem to want the work that feels like a person stood behind it.

By the end of the night something close to agreement had formed, loose and full of holes that people kept poking at, but real. What they had landed on, roughly, was that the designer’s center of gravity is moving. Away from what the interface looks like, toward what the system is allowed to do and how it should behave.

This connected to something Pablo had been turning over all evening. If the old job was building the layer of software between a person and a computer’s capability, the new one might be designing the framework the system thinks inside. It isn’t the interface anymore. It’s the logic of how the system goes at a problem, which questions it asks and in what order. “The design is going to be a framework of principles of how to think about something when interfacing with the technology. Controlled emergence is something I keep thinking about and creatively toying with,” he said, in a side conversation several people leaned into. That is a more structural way of working than designers are used to, and it is still design.

Saurabh Kakade, a Creative Strategist at Meta’s Creative Shop, described what this looks like in practice. His job is advising some of the world’s largest advertisers on creative, but the work he cares most about was never in the job description. “I went upstream, to where the AI systems themselves are being shaped, because creative judgment had to be in that room. Researchers know how to make models work. Engineers know how to make them scale. Neither discipline is trained to answer what’s worth making, or what ‘good’ should even mean, that’s what creatives carry. If we wait to be invited in after the technology is finished, we’re decorating decisions that have already been made.”

He raised Coca-Cola as a thought experiment. “Take a brand like Coca-Cola. India alone is so diverse you can’t treat it as one market, you can’t even hold Asia together as one idea. And yet a single brand guideline tries to contain all of it.” Now imagine dynamic personalization that works, where the creative and even the logo change from country to country and person to person. “That future is coming. The interfaces will be dynamic. The real question is, what ensures taste isn’t lost along the way?”

His own answer surprised some people. “Maybe the designer has to move closer to being an artist. Following guidelines and systems is the easiest thing for AI to do, so the human contribution has to come from somewhere deeper, from the soul and the origin of the work. You have to know a thing well enough to recognize it, or its absence, in a thousand generated variations.” “Maybe then,” Pablo offered, “the role now is to design the frameworks and guidelines for AI to follow.”

Kevin Funkhouser described it as a new layer opening on top of all the familiar deliverables. “The design of the soul of a product. Whatever that is. The morality. The definition of where we want the thing to play within. If at least you have a definition of what that boundary is, then sure, iterate and personalize away within this margin. And that’s almost what you’re designing against.”

Alex Prodaniuk called it curatorial. “If you can do millions of iterations, how do you decide what’s slop and what’s worth it? That soul, that artistry, it’s all about how good you can be at curating.”

Conner Drew made it concrete. At Monarch Money, the variance in how people want to manage their money is enormous. “I have never worked for a product that has so many people who want to do it this way, this way, that way. Creating a single product that could work for everyone is nearly impossible.” He sees AI as a way through that, but only if designers set where personalization stops.

More and more, the heart of the job is deciding what gets made at all. You set the boundaries a system runs inside, the rules for what it can and can’t do to a brand when no human is checking every output. Then you look at a thousand generated variations and keep the few that are any good. It is upstream work, and it asks for something a brief never used to: taste, and the conviction to kill work that is technically fine and has nothing inside it. People who can do this don’t get less useful as the tools improve. They get harder to replace. Their title might change, but the deciding won’t.

Near the end of the night, John Roescher put a question to the table. “Optimism is kind of core to why design exists. It can be better, and we are going to try to make it better.”

“Good designers are progressive thinkers by nature. Because we care and we have a drive to make stuff better,” Pablo added.

People answered it differently, but they kept landing in the same place. The tools will keep changing and the economics with them. Competent work will keep getting cheaper to make. Through all of it, some things will still take a person who has thought hard about what they’re making and why.

Peter Markatos put it in terms of what he wants for his 15-year-old son. The way he described it, a designer now needs to have lived enough to have real judgment about what’s good, and to be able to put that judgment into words. “Creative directors who are articulate and have a rich life and can actually describe things, they’re still going to have a job.”

What he valued wasn’t the latest tools or the fastest turnaround. It was a person who had lived enough to develop taste, and could put it into words and into work.

James Song, who had spent the conversation refusing to treat any of this as doom, landed on the least comfortable point of the night. “Given all these cultural, economic and societal pressures on people these days, I don’t think ‘craft’ is enough,” he said. “You’ve got to find a way to make something that matters for people, even if there’s no UI at all.” And the job that is left, once the screens and the systems are someone else’s problem? “To make sure that people feel great about making decisions with your product.”

That’s a high bar, and it always has been. AI just makes it easier to see. So even with everything changing around design, what sits at the center of it may not be changing much at all.