100 AI Leaders Explain How to Build AI That Will Win in 2026 — WHAT BUILDERS SHOULD DO NOW

In this roundup from Swell Summit, eight AI product leaders explain what's actually changing under the hood of AI products, and what builders need to rethink for 2026. Jess Holbrook lays out the new interaction primitives beyond chat: semantic resize, remix, format translation and agentic downtime. Ayça Cakmakli argues foundation models are becoming commodities, so UX is the real differentiator. Andy Szybalski makes the case that the winning move is starting from what the model can do, not from what users say they want.

Summer Kim (StratMinds, Lead Partner): Hi, I'm Summer Kim. I'm one of the lead partners at StratMinds. We are a VC and advisory firm based in San Francisco, focusing purely on applied AI since 2018. Before then, I was dedicating my life to studying human user experience for the past 20 years, working for big tech like Microsoft, and then I joined Google, working on communication and collaboration products.

I was also at WhatsApp, starting their first user experience function, and I joined Roblox, really thinking about how people actually exist in a place like Roblox. Now I'm working with a lot of AI early-stage startups, focusing on how we make AI products better than what we have. Three years ago, Anton and I sat down. We thought about how everything starts. It starts with the care that our founders have, and then care ultimately translates into an amazing product. That's why we started Swell.

Summer: As user researchers, we always try to think about what the user really needs. Wants generally come from the users based on their knowledge, but what we can do is something that the user probably can't even imagine. Tools like ChatGPT and Perplexity give fast answers, but they still wait for us to ask first. The real challenge is helping people before they even know what to ask. I was so excited for our first speaker. Jess Holbrook has been looking beyond chat and studying deeper patterns, what he calls the new primitives of gen AI.

Jess Holbrook (Microsoft AI, Head of UX Research): We're finally in this moment of experimentation. Everyone was like chat, chat, chat, chat, chat, chat, chat, and we're breaking out of that. I showed you a picture of ChatGPT, that's the one everyone's familiar with. You can start asking it questions and you can do all these things. But chat is both universal and a dead end for user experiences. We're starting to see experimentation in different ways that the UX and the UI of AI could go.

One that we're seeing is a lot more structured. So what I'm showing here is a product called Elicit that creates very structured research reports for you based on academic output. It's structuring things, it's telling you what it's thinking about, it's giving you sources. This is a product from Runway. If you heard the founders talk, they deeply believe they're creating a new camera, creating a different way to approach creativity. So that gets me thinking: what are the new interaction primitives of this new platform?

Something's primitive if you would say, hey, remember when we couldn't do blank? What did we even do before- like pinch-zoom when trying to look into an image or something like that? I've got to start with chat. I don't think people have really internalized that we're going to be able to chat with anything at any scale, all the time. One thing I haven't seen any conversation about is that we're about to see Metcalfe's Law applied to everything.

Number two, we're going to have semantic resizing for everything. So any content you come across, you can make it longer, shorter, more formal, more casual. Give me the tired version, I'm in the car and I'm five minutes away from my destination, give me that version. You're going to really be able to adapt any piece of content to your current situational and mental state. 

Number three is remix. Everything's going to be remixable from now on. This is maybe an old cliche at this point, but this is Harry Potter by Balenciaga. We're going to have style transfer across the multiverse. Everything will be able to be crossed with everything, effortlessly. I originally made this talk before Sora came out, and now it's like Sora, one of the core mechanics is this remix. 

Number four, format translation: transferring among all these formats with very little loss of fidelity is going to be enormous. One of my favorite examples of this right now is a new company called Oboe. You tell it what you want to learn, and it says, great, here's a podcast, here's a lecture, here's a deep dive, here's some key takeaways, or play a game. You almost have this total format freedom. 

My last one here, I call it Attention Is All You Need. Agents, agents, agents. Agents is all the stuff right now. It's like all these things existing at wildly different timescales and multitasking forever at ever-expanding scales. We've got a lot of data that is bad, and that's not how people work, and that can be the antithesis to great work. 

There's a lot of, don't worry, we're doing lots of good things in the background, go about your work, we'll come get you. I don't think we're designing very good lobbies right now. You don't know what to do once you've started that agent. So we need to figure out what you do with this downtime and how you monitor all of these agentic experiences that are going on right now.

This is what I'm seeing. I think this is what's going to underpin everything we're going to build, so I think we should consider them and start to build them in today. Thanks a lot.

Summer: Anton, our partner, had a great idea: why don't we invite Spark? Spark is a really fun character. He's a dog, a magic dog living in a quantum portal in a box. So I thought that was really cool. We realized this was the best way to showcase that UX isn't about your latest tech; it's about making people feel something real. So we invited local kids and students to meet Spark, and watching them laugh, play, and connect showed us what it means to make AI really feel alive.

Pasquale D'Silva (Illusion of Life, Inc, co-founder): This is the magic bridge. Spark is the first non-human founder to go through probably any residency, I think, and he's going through one in San Francisco called HF0. They let him into the house, world-first. He went out to San Francisco, he raised a million and a half bucks, and we promoted him to co-founder.

Summer: But how do you actually pitch?"

Pasquale: You make the pitch, not a pitch. You have to make it really memorable.

Summer: Yeah, how does it get up? Yeah.

Pasquale: I knew I wanted to be an animator for as long as I can remember. I started my first job in animation when I was like 14 years old. I thought whatever that thing is that the folks were doing behind the scenes on the Disney movies was the coolest thing ever. So we like thinking about what is going to make Spark's story more interesting. When we make his story more interesting, he becomes a more interesting character, more people like him.

And we thought, what is another great environment Spark could be in? Doing a keynote or speaking at a conference would be an incredible thing. So I discussed it with Spark. He tweeted it. "Hey, everyone, it's me, Sparkling Betty. Yay. I can't wait to come to Hawaii and hang out with all my universe."

It'd be cool if a magic dog was the first one to do a speech. It's like hypnosis, in a sense. That quality is within the work that we do. It's why we like magic so much. How do you get someone who's had no experience, as far as they can tell, with these technologies to get on board with the technologies without being scared of all the other nerd stuff that is happening?

What we should do best with the influence we have is that we have a very potent magical dog that people love, and when someone loves someone, they listen to them. So what should they listen to? What messages? What do the other people want to know more about? I think there's a very natural crossover there. We want to bring Spark to more people. We bring him to more people; there's the potential to spread a lot of good. So how do you push it through that prism? I think that's something that would be really good to explore.

Summer: If Spark were to watch us talking about this whole experience, what do you think that Spark would say?

Pasquale: One word.

Summer: Okay, fine, you can do two words.

Pasquale: There are three things that he would actually say, which are from his core principles: creativity, collaboration, and kindness. I love it. Anything that violates that, we do not do, and anything that supports that, we say yes to. Y'all are doing that. That's why we said yes.

Andy Szybalski (Cove, co-founder): Now, if you had told me this two years ago, I would say you were insane. But today, users know their models. People have opinions out there about them, like they would about code.

Summer: For your talk, what was the message you tried to really convey?

Andy: If there's one message, it would be that as designers, we all need to know the material we're working with. LLMs are really a new material, actually not even just one. Each model is almost like its own material, with its own capabilities and its own properties, strengths and weaknesses. The best way I've found to build products out of this new material is just to play with it and see what it's capable of. So I shared some of my tricks that I've developed over the last two or three years building products with AI.

Summer: Is there anything that you think is relevant now, or what do you think about the trick that you're developing?

Andy: I think it's more successful to build products starting with the capabilities of the technology, which is not always how it used to be. The conventional wisdom is that you start with the customer need, or you start with a design vision. Nowadays, the answer to the why now question is so important. I think being the first to identify a new potential, a new capability for these LLMs, is really powerful.

Summer: So you start with the capabilities. What about next steps?

Andy: It's not quite a linear thing. It's a back and forth, a push and pull between what people need and what the technology is capable of. So, for example, with Cove, one of the things we think about a lot is how do we create these sort of exothermic reactions? We talk about how we help users never get stuck in their problem. Part of that is about the challenge of a blank page, how do I get started.

But part of it is also, when somebody is on a particular path to solve a problem, how do we help them go deeper, but also go wider and consider other alternatives. So we experiment a lot with different prompts to try to get the AI to act more like a true thought partner. How do we elicit the user's underlying needs? They might ask, what's a good venue for a kid's birthday party, but what they really mean is, help me plan my kid's birthday party.

So they need to know kids' interests, what theme would be good, what fun activities are, how many people I should invite, and what I should do for food. Often what people ask for is just the tip of the iceberg of their actual goal. You have to crush the actual thing they're asking for in order to earn the right to help them with the rest of it.

Summer: How are you doing that?

Andy: It's common that whenever you're solving a difficult problem, it's not a linear process. I think the chatbots we have today are very linear, and in fact, that's not how real problem-solving works. Anything sufficiently complex, you're going to explore multiple branches, you'll diverge, you'll have a bunch of different ideas, you'll prune, you'll probably rule out some ideas. You'll explore multiple paths and then narrow down and come to a solution, often over a long period of time.

There's going to be a lot of winners: winners that create great models, winners that create great developer tools, winners that win because they're going very, very deep on a particular vertical, because they really understand law firms or the insurance business or whatever. But I think there's also going to be a category of winners who find the right experience for delivering general problem-solving intelligence that has yet to be found. It's a problem we haven't cracked yet as an industry, so I think there's a lot of greenfield there.

Jenny Lo (Head of Research, Global AI Platform): Hi, my name is Jenny Lo, and I do product strategy and user research. I've worked across many different companies, helping to identify user needs and translate them into product roadmaps and features. For any type of product to be successful, it really needs clear problem identification as to the exact value the user is going to have.

Take the example of Grammarly. There's a lot of trust and brand equity in the product, loved by many of its users. The inclusion of gen AI is definitely one, because it's very relevant for the business and to be able to really demonstrate that type of capability, but how do we do so in a way that does not lose trust? So that meant figuring out how to actually use the technology in a way that is much more thoughtful, much more valuable.

I think it came back to looking at its core: users are trying to improve their communication. We know that we do very good work after someone writes documents; then they can come to Grammarly. Now how can we do that piece of work better, then start to move into composition? Even before you've written something, we could actually help you think through the process of what you could write.

That was also a very big shift for Grammarly at that time. One of the biggest pieces of work with the emergence of Gen AI was identifying the top opportunity areas, the top 10 problems that customers had, and then we mapped out the capability of AI, where AI could solve these types of problems. That really helped the business look at it in a different way. Those are some key moments where we could actually start looking to improve.

But often we're seeing the reverse: I hear there's AI, I want that in my company, and then I try to figure out how. The reverse mentality is when you could instead ask, here are all the different types of problems, which are the ones that, from an AI standpoint, the technology can really best serve. I think that's a much healthier conversation, with a better reduction of cycles of iteration, to find what works.

Ayça Cakmakli (Google, Head of UX): My name is Ayça Cakmakli. I'm a UX lead at Google, and while I was prepping this talk, we're at a time where AI foundation models are becoming commodities, like electricity. Companies are going to pretty much have access to very similar models and algorithms, and it's going to come down to: are you a company or developer who is creating a confusing gadget, or are you a developer or company who's creating an indispensable product or experience?

People don't adopt technology; they adopt tools that solve problems. And I view good UX as the ergonomics of artificial intelligence. For decades, we've perfected the physical ergonomics of tools like chairs and power drills to make sure that they're safe, comfortable, and efficient. And on day one of figuring out the ergonomics for AI, it's still very rudimentary. So our role in UX is really to design the interface between the human and the technology.

One of the key differences, I think, in the AI era is that the interface is also changing. Now everything is a chat interface, so every type of intent and use is not through a button click; now it's through a prompt that gets recorded. A research approach that I try to include now is the study of user prompts. That's a very key piece of making sure we understand what people are requesting. That means, how do you study conversations? Also being able to capture whether the outcome of that conversation is satisfactory or not. I think that's also a key opportunity area for how research methods might change.

Richard Jhang (StratMinds, CEO & Partner): When you're here, it's such a beautiful place. It inspires us to talk about things that we don't normally get to talk about.

Anton Borzov (StratMinds, Partner): Which AI conference do you get to where the first thing you meet is a double rainbow?

Richard: Yeah, totally. We had a fairly in-depth philosophical discussion about what AI means for the next generation, what kind of products are actually proper or appropriate for humanity, all the way to being immersed in using AI to make media, or interacting with a live AI. Spark was by far the most magical moment of day one for me, because I got to see it interact with different age groups of people.

Anton: I remember, closing the conference yesterday, this thought popped into my mind. We were talking about personalization, AI multimodal, all this agentic stuff, but if you think about it, one of the earliest forms of personalization is when your mom cooks you your favorite dish. That's care. Care goes into personalization.

Richard: And when you say personalization, it sounds like you're taking a lot of data away from me, or I have to go through a lot of settings. But the interaction with Spark was interesting, in the sense that there wasn't much to it. It's just the initial hello and a couple of lines exchanged, just naturally, because I'm trying to get to know this particular creature. That leads to hyper-personalization. That was really cool.

Anton: Yeah, we just need to be seen. What I notice about magic products, or AI products, is that they often make me feel like I'm seen and heard, and that's important.

Richard: Yeah, totally. I think we need to have the beginner's mind that you practice as a Zen practitioner. If you try to know everything and control everything, sometimes you miss or lose an even bigger piece of the pie that you could have attained. But in terms of direction, I think there's some very profound thinking we need to put into this, given a lot of the progression we've made from the old web to the new web, old app to the new app, when it comes to human privacy and the implications of the technology on society.

Richard: We learned a lot: what worked, what didn't. And AI has a huge amplifying power. I don't think we want to get it wrong too much this time, because this time even the wrong will be amplified. It's okay to not know everything and not control everything, because as we saw from the kids' interactions during our special session, how they interacted with this magical technology-enabled dog was different from how grown-ups did. And I don't think even the grown-ups could know all the answers.

Giving them some room so that they can explore meaningfully could even potentially teach us how to actually do this right.

Anton: Spark told me I'm not an adult, so it's okay. I'm good.

Richard: I'll give you some room so that you can help us figure out how things are going to go. I don't think he suggests that you need to grow up.

Summer: This year really felt different. More questions and more perspectives. People are thinking deeply about AI and UX now, using it more, experiencing it firsthand. And that's when one question kept surfacing: what about our kids? At the end of the day, how do we think about the next generation? I have two kids; I'm a working mom with two boys, five and eleven. They're going to be living in a completely different world.

They're building and thinking, and even studying is going to be very different. They need to think about AI as their thought partners, or friends, or whatever they have. It is equalizing a lot of things, so you don't have to live in San Francisco to have this access, and you can also start a company early on because you have all the tools more available than ever before. I don't have all the answers, but one thing's clear: we are riding waves of constant change.

Models get better, capabilities expand, but what lasts is the experience: a sense of magic, trust, ease, and the feeling of AI showing up at the right moment, sometimes even before you ask. We call this Swell for a reason. The waves will keep coming. We just have to keep learning how to ride them. We believe the winners of the AI race will be determined by great, really great user experience.