In this interview for the First of Kind series, Soleio talks with Ryo Lu, design lead at Cursor, about how design and product work are evolving in the age of AI. Ryo explains that Cursor’s design team is tiny (four people) and that most engineers also contribute to design, making the role of a design lead one of "wrangling" and unifying chaotic, code‑driven ideas. He describes the hiring philosophy of hiring generalist, high‑agency individuals who blur the lines between design, engineering, and product. The conversation delves into Cursor’s recursive dog‑fooding: the team builds the tool with the tool, creating a self‑improving loop where personal pain points become product improvements. Ryo outlines the shift from low‑level code editing to higher‑level agent‑orchestrated workflows, noting that humans become the bottleneck as agents become more autonomous. He connects this to a broader design philosophy of a minimal pattern language – a set of Platonic building blocks that can be composed into emergent complexity, a concept he calls "vibe coding" from his earlier work. The interview also covers prototyping environments like "Baby Cursor" that let designers experiment without production constraints, and the idea that code is the unifying medium that can accommodate visual, narrative, and functional inputs. Ryo reflects on AI’s dual effect of leveling the playing field while raising the ceiling for experienced engineers, and he argues that the future designer will be a multi‑modal fluency practitioner – a "vibe coder" who collaborates with agents rather than merely creating static mocks. The discussion ends with Ryo’s open‑ended vision of solving software and seeing where the journey leads.
Chapters
If you're a designer who thinks designing is just drawing buttons and inputs and making mocks and maybe like building design systems, you're like a little fucked.
First of Kind is a new series about design and tech told by its pioneers, sponsored by Framer, Cursor, Vercel, and Profound. Rio Lu leads design at Cursor, the AI tool that revolutionized how teams build software. He was vibecoding long before the term even existed. This is our conversation. Rio, how are you?
Good. How are you?
Fantastic. Thank you so much for spending time with me. Yes, I so appreciate it. You and I have very similar career paths, even though it's very different contexts. And when I look back on my career, getting to work closely with Mark Zuckerberg on a generational company as an early designer, and then moving on to run design in my own way at a company called Dropbox years and years later, I couldn't help but see some parallels with some of the work that you've done at some transformational companies, working closely with Ivan on Notion in the early years, and now leading design at Cursor, probably one of the most, if not the most influential startups in the age of AI. I wanna talk to you about your path to here.
Mm-hmm.
Because the whole spirit of this program, First of a Kind, is people who are doing something for the first time, the, the first of n. And when I look at the space of designers working in our field, you have consistently shown that you are an original. And I want to talk a little bit about winding the clock back. How did that come to be? Like, what is like the life story behind, uh, the person who's leading design at Cursor? Let's start with, uh, Cursor today, and then we'll walk backwards. What is design at Cursor?
Design at Cursor right now is 4 people, including myself. 2 people focus on the product, 1 person on the brand. But a lot of, like, almost all the engineers at Cursor are also kind of designing, or like people start building stuff, and as they build, they're like actually designing the system. Maybe like they, they care less about the visuals or how things flow at one point. Yes. But they're still thinking about it, they're building it. And then now my job is how do I kind of wrangle things back? How do I clean it up? How do I unify the concepts so that things stay cohesive. It is a lot more chaotic, but that makes it more fun.
Interesting. Do you find that a lot of the work you then do is curating the ideas and trying to find their platonic form? Like, what does the day-to-day look like given the size of the team, the different skill sets?
Yeah, yeah, there's a lot of like, you're both thinking ahead, like, Where do we want to get to given what people are doing today, the systems that we have today, but also you're like exploring new ideas. Maybe they're a little crazy, a little further out, maybe they're like closer. And there's all this code that gets produced, that gets merged, and sometimes maybe we spend more time thinking ahead than we know maybe a little bit more clearly at a high level, how should things fit together better. But also sometimes you're just playing with some ideas and then see how it works. Once it's built, then you have better, clear, like a better idea, and you keep going. And that's good too. But a lot of times, because we're working with software, a lot of the concepts, a lot of the systems part of it, it's all the same thing. So I help people like break down the problem, see, ah, this piece and this piece is the same. Maybe you should talk to this guy because he's working on that. Maybe your thing and his thing is the same thing. Maybe we already have the thing, we just need to make that thing better. Maybe we already have the thing, we just need to cut it because we don't need it anymore, right? And then you're trying to keep the system clean. At every layer of abstraction. Yes. From how people perceive it, like visually, to like how it's modeled, to how, you know, things scale, to what are the core ideas. And they all kind of are tied together.
I imagine that this way of working shapes like the kind of people you bring to Cursor, you hire. What are those people like and how are they different from other startups? What do you look for to kind of tease out their qualities.
Yeah, I feel like a lot of us are like generalists, super ICs, ex-founders. Like, people have a lot of self-agency and skills to kind of just do the thing, and they're like less worried, and they're good collaborators. They don't care about boundaries or process or bureaucracy. We have this big problem. Maybe I'll focus on the parts that I care most about, that I'm good at. Maybe for myself, it's like the concepts, the visuals, how things fit. The engineers can think about the architecture and stuff. And then we used to not have PMs. Now we do, but they're like They just found, you know, there are certain things that I'm good at, so I just come and then we work together, together on the same material, which is code. And it doesn't really matter. It's like the old ways of doing things is more each discipline does their own thing in their own little world with their own language. Tools, and then you try to stitch things with like artifacts and processes. Like the PM writes a PRD, runs meetings, shares the things out. The designers mix the mocks, the pixels, with the concepts that maybe that is not theirs. The engineers take whatever that's broken down from all of this and then just do little tickets. When they should probably think about like, how do I architect it better and, you know, make it scale more. They should also have some fun. So all of these people are so disjointed. They all use separate tools, they all talk differently. What the fuck are we doing? Versus we're all like software builders. Some of us might be good at words, or pixels, were like actually really good at coding. And you just put these people together, give them like a big and kind of vague problem. Might be seemingly impossible if framed in like a more traditional tech company context. But then here it just happens because you give these people agency.
Yeah. What I find amazing about what you just described is that Cursor today is becoming more and more the surface for creation. And in this context, you are building Cursor within Cursor. There's something intensely recursive about that, right? And so it creates this like quality of dogfooding that's very unique. To people who are unfamiliar with that way of working, how do you kind of describe it to yourself? I don't know how you describe it to others.
Yeah, I think It's like a lot of us, maybe we think of us as we're just like doing software for a job versus you're actually like, you're building something that I really care about. Like, what's the difference? Like, if you want to do something you really care about, you just need to be constantly in it. You start from building things for yourself. Your own pain points, your own habits, to make yourself happy, to build something that flows well, you know, with how you think. Then it gets, you know, a little bit bigger. It's like now you're solving problems for people like you. In the context of Cursor, everyone who started Cursor, they were just, you know, software engineers. They love programming, they love making tools, and that's how they started. They needed something that fits the AI better, that takes advantage of all the things the models have, put them with coding, like what is the right interaction, what is the right way to do the thing now. They started there, they built them, you know, built the thing for themselves, and then they kept doing that. Because they care. And now you get like more people around who also care. And you don't treat it as I need to wait for the ticket to arrive to get assigned, then I do something. Once I'm done, I don't care because I can't really use the product myself. I don't, I don't, I can't make the connection even when this thing is like something so core to us. Like we use it almost every, like every hour, every second, both to make Cursor and also like just to have fun too. Like a lot of us, we started from building like side projects. So like one of our engineers, Ian, he used Cursor to build like a 3D like Minecraft thing in the browser. I used it to like make my own personal OS just for fun. Just to play with it, just to feel it. Then we use this feeling, we use all the things we observe that we hate about the thing to make it better. And you just keep doing it. So the tool gets better, you also get better. Like, you're now more capable, you can do more things. And as you do that, you improve the tool, and it's like a self-improving loop. Yeah, once the agent is able to do this by themselves, it will be even faster.
Yeah, and do you feel like you're very close to that threshold? Because from my vantage point, the entire tech ecosystem is in the, like, the drag field of Cursor. This, like, recursive flywheel that you described, yeah, it's very palpable. Teams are shipping faster everywhere. Engineers that lean into the technology are wildly more productive than even years past and nobody wants to retreat from that productivity. What will be the shift in self-improvement once agents become a core part of the overall flow and you're now starting to coordinate?
Yeah, I think we're gonna ascend in like the layer of abstraction we interact with these things, like with the agents or even the code. The software itself. Like we started from manual coding files, lines of code, typing lines of code to you type and then we start, we complete the function for you. We figure out what's the next thing you want to write. We help you jump to that file. We let you keep in your flow as you type. Then we did the chat thing. Now the AI is aware of your whole code base. It can answer all the questions. We added editing capabilities, it can now apply changes. We added tools, then now it's like more agentic, it can do more things. But we're still limited by ourselves as humans. We can't process that much of information at once. We can only look at one chat thread at a time as it happens. Maybe like the most crazy people I know, the best programmers, They can maybe only do like 4 agents max at once. Then the problem becomes as the agents get more autonomous, they get more context, they have more tools to do things. They might also know what you want more clearly. So like I can capture my high-level intent and plan. Then the agent can do a lot of things, but you become the bottleneck. How do I help the user instead of like chatting with 15 agents at once? Or reviewing changes from all these things. Like what is the next level interface or the abstraction?
Yeah. And does that become the thing that unlocks a concentric circle of new users?
Exactly.
All of a sudden now you have—
Now you ascend in the abstractions instead of directly interfacing with code, you chat with it. Instead of directly chatting with each agent, maybe you plan with it. Maybe you see things at a higher level. You see a dashboard of what, oh, these agents are kind of blocked, I need to help them. These are the plans and tasks that I want to do. But maybe, you know, just maybe the designers want something else too. Like they want something more visual, directly like manipulable. Click on this, make it bigger, boom, it's bigger.
The thing that I've always found kind of remarkable about Cursor is this like larger theory that I feel like you have been carrying since before your time there. This idea of like a pattern language that, um, is extensible, that you focus on, on, on the Platonic forms of the building blocks and you, you honor them and try to really get the most out of them and allow that their composition to be as infinite and wildly useful as possible so that the system, um, can almost telescope up and down in complexity. You can have a set of keys that can fit the hands of a child, or you can have a set of keys that can fill a room and play for an auditorium, right? And in my mind, the thing that I find so difficult about this job is that what you're describing makes sense to me, except these models change their powers, their capabilities, and there's a feel to it, right? And I feel like before we had vibe coding, you were feeling the shift in those models. How do those two ideas intersect? Because I feel like it's at the heart of the design philosophy that you've carried from your time at Stripe and then at Notion and now at Cursor.
Yeah, I think pre-AI, it was mostly us thinking about what are the parts, what are the things like the comp— the, the lower level components and how to kind of assemble them. If you look at each piece of software at the conceptual level, usually there's not that many things in there. Say like TikTok is just like a feed of videos auto-looping. The user, all they do is they go to the next one. Notion is blocks, pages, databases. But once you kind of combine these things, it can give rise to like emergent complexity. People can do whatever. The simpler the core set of things, the easier they can combine. When you treat, say, like a lot of people when they do products, they're thinking, ah, this is my group of users that I care about. I'll figure out what are their problems. Maybe I found 20 problems, but I decided to focus on one problem and I just make one thing. Say my concepts entirely are based on that one problem and I make my whole thing out of it. It just won't scale. It won't be as flexible. When now it's like with AI, because the models are themselves general purpose, they have read the whole world, all the public information, knowledge, all the repos you see on GitHub, then the thing itself, like, you know, truthfully it is not something specific just for this group of people, just for programmers. Just for coding even. The tool should ideally fit you, how you think, your preferred input, like input method. So like, we as designers, we love pictures.
Yes.
We think in 2D space. Sometimes I draw like little boxes to convey my ideas. Sometimes it's like a high-fidelity mock. Your PMs are just like doing their specifications, setting the constraints in text. But longer, not like a short chat prompt. And then the coders, maybe they're like, they're also building some prototype, trying their idea. You can actually unify all of these different inputs. And then you can also create an interface because we're like an interface that adapts to people that is not stuck in what the designer had in their mind for that one specific problem. The interface can adapt, it changes based on what you do, based on who you are, if you're a single person, if you're working with a team, with a big team, the kind of things you do, the piece of, like, the piece that you care about, the form that you prefer. Like if you like the terminal, you like doing CLI stuff, that's fine too. I think it's fine. Or maybe you're on the phone, I'm just checking out like I have some random idea, I want to kind of just try it out. That works too. But they all talk to the same agent, the same underlying structures.
And Cursor becomes this lingua franca that sort of exists almost as a new type of state machine, some sort of new surface in which people with different languages, different ergonomics can work on a common system and common sets of agents supporting that system.
And we're kind of already doing that right now. But the world doesn't give us like one thing that ties everything together. We're all speaking with our own separate tools and agents.
Right. What is it— what is going to be the unifying element there? Because in some respects, part of human nature and human folly is that we create pattern languages to explain the world. And I love the examples you gave for like the PM with the PRD or the designer with a canvas tool. For me, like, I love analogies. I feel like analogies are like, yes, another form of ergonomics. Sometimes you can arrive at a deep understanding of something with a really good analogy. You can also deceive yourself with a really good analogy that's poorly applied. What is the unifying substrate then that cursor is going to start to enter? And what is the kind of designer or the kind of design team that you need to build or assemble to explore that opportunity space and find the answer?
Right. Like, I see the unifying medium as code. And you're trying to kind of draw like arrows there. Instead of them thinking, oh shit, I need to learn this completely new thing and then figure that out. How do I kind of keep my habits, keep my strength, all the years of Figma-ing that I've done? Like, don't throw those away. But with this new thing, you can be like 100 times more productive. You can do a lot more than you were doing before because the agent kind of covers the parts that you might not do well, or maybe you're just starting to learn.
One of the things that I feel has made you a first-of-a-kind designer, there are these smaller environments, and there's something, um, almost, uh, cosmological about it that I love. You've created Rio OS. It's like a, as a working example, but really as an environment in which you're using cursor, feeling cursor.
Yep.
Seeing its capabilities in practice, but also in some ways like inspiring Cursor, right, in a new form factor. You've built mini Cursor.
Baby Cursor.
Baby Cursor, that's right. Baby Cursor. You've built Baby Cursor, which is like a different representation of the same system. Uh, it's like, in some ways it feels like a video game inside of a video game. Yeah, yeah, yeah. That informs the larger video game. And so there's, there's something I call it cosmological because I feel like you're now approaching some deeper truth.
Yep, yep. You realize all of these things are actually the same thing.
Correct.
And like the essence of it is the same, but then as you— the environment kind of sets the complexity.
Say more about that.
You know, the meme is we're a VS Code fork, right? So you're already in someone else's world and there are certain things that are hard to change. Then you need to kind of fit yourself in there, figure out a way to do that. And, you know, sometimes it is hard to say, I just have this idea I want to try. But now I need to spin up my backend server, the API, the client, I need to start all the scripts. And then the thing that I will make, I don't want to care about all of this. All I want to do is to just try my idea. So you build a simpler environment for that idea to emerge. So I built baby Cursor that looks exactly like Cursor, but it might have parts that are closer to my ideal. It might feel really real, but then it's actually not.
Right.
It is just a prototyping environment for more ideas to emerge. And we noticed this happened not just at Cursor. I talk to people at like Ramp, Notion, they're all like building little prototyping environments for the designers to just have fun and play with ideas there. They don't have to care about, uh, this code has to be, you know, merged into production, wrapped in like some feature flag and whatever.
Right. Yeah.
Yeah. Less of that.
Do you think it just ends up being like a software version of The architects creating models of buildings?
Yes, exactly what that is.
Okay.
And then the problem right now is, or like even say with architecture or any physical things, the model can't turn into the real thing. But this time it can actually turn into real thing.
Yeah.
We haven't solved that yet, but I feel like that might be a way for people to say build software by building like miniature versions where You know, starting from the prototype, turning the prototype as an input into the actual thing, you know.
Yeah, yeah, we're coalescing.
Yeah, yeah. And if you look at like history back in the days or the earlier days, nobody like thinks of themselves as, ah, I'm just a designer, right? Or I'm just like a programmer. I guess everyone were like a programmer in some sense. They kind of do the whole thing.
So then what will be the shape of the new designer in this paradigm? Because I feel as though, like, again, I think of you as the original vibe coder. Yeah. When that term was coined and took root, I know who that is. They're describing Ryo.
Yeah.
It's him.
It's like, if you're a designer who thinks designing is just drawing buttons and inputs and making mocks and maybe like building design systems, you're like a little fucked in a sense that I think even like the models today, they kind of solved a lot of the problems on say implementing these things or even like now the image models can come up with like pretty good mocks. Yeah. Then what do we do? Then what is important? What's important is actually you need to help people around you and help yourself, like, see the thing in this, like, truest form. Both, like, say, for now in the present and in the future, what could it become? What are the things in there? What are the parts of it? Which parts do I need to change? What do I need to tweak?
Right.
But that happens also like at every layer. Maybe you're tweaking the visuals, changing the brand perception, reorganizing the product concepts, updating the flows to reuse the new patterns that we created. And you need to kind of do it, right? Yeah. So you might do it through storytelling, through code, through pictures. Depends on who you work with, right? Like, some engineers that I work with, they just love mocks, so I make them. But some people, maybe they're like, they actually just want to explore themselves, come up with a few options, like prototype themselves. Then I'll be like, this one is cool, this one maybe we should merge them. We keep, keep revving on it.
I, I really want to drill in on something that I've been thinking about a lot. Which is, I used to think that technology specifically created this egalitarian push of power. It empowered people to do things on the edges. But with AI, I've observed that there's also this simultaneous concentration of power where like the 10x engineer is now 1,000x more productive. And it's almost like a power shift to people who are building in these new ways. Often empowered by cursor. And I want to know, like, if you feel that tension, or if you feel as though you need to make a commitment to one or the other, right? Because the answer seems to shape a lot of, like, the product philosophy and how you think about— right?
I think it's more like you both kind of level the playing field and you raise the ceiling at the same time. Say like before, it's just like impossible for you to do anything before you like learn all of the programming concepts and the syntax, how to like run the server, da da da da. But now anyone with an idea can do it. Though maybe like the first output is a little sloppy, like purple gradients, but if you work on it more, iterate on it more, As you see the results, you can make it closer to what you wanted, but there's always kind of like, like you might not be able to know how to exactly instruct the agent with the exact thing you wanted. And that still requires like knowing what's there, knowing the language, the concepts, being able to read code. See the diffs sometimes even. Like the more, say, serious, or maybe you're just playing, it's fine to not look. When it's like used by millions and it's like critical, you have to. But as the agent gets better, it will also kind of help you do these things. Then more people get leveled up, but I think, you know, for people who have done this for years, who have accumulated experience, they should also like do their work more in flow, like extract out their strengths, multiplying it by like 10 or 100, combining it with other, say, more special— like specialists or professionals who maybe care more in depth in different domains, merging all of that together. Then collectively you make better software. Boom, you're like already up here. You can do things now. The more you do it, the more, you know, the deeper you learn, more capabilities you get, intuition about the models and how to do stuff. I like— I noticed some vibe coders actually do the agent stuff better than some of the more manual coders who haven't, you know, caught up yet. Yeah, because they've built so much with it, they've formed some intuition on, ah, this is how I should like break down my prompts, and this is the size of problem this model is good at, because you have to build, you have to use it to, to learn that.
It's a different fluency. Designers often ask me What tools should I be learning in this new AI paradigm? And we know all the current tools and the new ones that are emerging, but maybe the answer is Cursor. Why is the answer Cursor from your perspective?
I think a lot of the other tools are maybe a little too opinionated, too siloed in their layer.
Oh, interesting. Say more about that.
So like Figma, You're like stuck in the pixels. Maybe the V0 lovable, they're also kind of boxing you up in the safe area, whereas like if you do stuff within that area, it gives you, say, maybe slightly better output. Yeah, but once you go beyond that, you can't. But Cursor, how I want to kind of shape it towards— I don't want to force you to do things in a certain way. If you have found your way, you can keep doing it.
Yeah.
If you haven't, you can find your way. It's like, I don't want the tool to be stuck as a code editor because it's no longer—
before this conversation, I ever thought of Cursor as a, like, a spirit inside of VS Code, an idea that's kind of emerging. But I now feel like I have a different perspective on it. It's the thing that kind of becomes the metal that we try to, try to stay close to when we say stay close to the metal, and that it doesn't necessarily have to conform to one set of ergonomics. It can instead just meet the person where they are, and that seems to always be like the recurring theme in the design work that you do.
Yes. I don't want to force people to change. I want them to figure themselves out. It's like, I've lived in China, Canada, here. I've seen how people do stuff or how people behave, and we're all different but very similar. Like, each of us have our unique ways of seeing the world. Yes, we're thinking, we're telling what we want. I think the best way to kind of like pull the good stuff out of the AI is so that you can express yourself in the way you do. You're not forced to learn another language. You're not forced to study my concepts. This is the way to do things, because that's never true. The way changes, and it's unique to you. And even like, say, when you're just doing stuff yourself versus you're doing stuff with your team, it changes. People find their way together.
Where do you think the way leads for you? Because when I was running design at Dropbox, I was not imagining myself as a full-time investor or a father to a teenager, you know, there's a— you have murky ideas of where you might end up. When you sort of look at your golden thread through the present, where do you hope it takes you? What are you building towards?
Solve software. Maybe just with this. Then I'll figure out what's next. I'm not sure.
Super. Well, I appreciate you leading the way by example every day in public and inspiring me and so many. Thank you. Thank you. This is First of Kind, a new series about design and tech told by its pioneers, sponsored by Framer, Cursor, Vercel, and Profound. Follow us on X, YouTube, Threads, Spotify, and Apple Podcasts. The future belongs to designers who build.
