Ryo Lu (Cursor): AI Turns Designers to Developers

0:00Over the last, I don't know, 15 years or so, the art of making software fragmented a lot. and then we kind of split into different roles. Each role kind of used their own tool, used their own artifact. They think in their own kind of words and lingo. With Cursor, things kind of flip again. For the first time, that design is such an approachable concept and skill set to a lot more people. And it brings together sort of people who have aspirations for design and wanting to build things, wanting to prototype things, putting beautiful stuff out in the world. There needs to be something for the human to specify what is good, what is right, how I want to do it.

0:43If you don't put in that opinion, it will just produce AI slump. People will always have their strength or their unique, special skill. I see AI almost like a universal interface. So design is kind of like trying to figure out what is the best configuration and the simplest state for all of us. The beauty of it is actually putting things all together.

1:12What happens when the designer becomes the developer? When mock-ups that used to die in Figma can suddenly become living products in minutes? My guests today are Jennifer Lee, general partner at A16Z, and Rio Liu, head of design at Cursor, the AI code editor that's collapsing the traditional boundaries between design and engineering. Rio spent years at Notion and Asana, watching his designs get stuck in endless meetings and handoffs. Now he's building tools that let designers ship real software themselves. We're exploring how AI is ending the era of fragmented teams, why taste isn't really worth talking about, and what it means when a single person could do what used to take an entire product team.

1:50Rio, welcome to the ACNC podcast. Jennifer, you've been thinking a lot about sort of evolution of design, evolution sort of as it relates to infra as well as software development. Why don't you talk about what got you so excited about having Rio and, you know, have this conversation? Rio and I got to know each other over the past few months talking about how large language models and AI tools are going to impact not just designers, design engineers, and how people are building prototypes and coming up with great ideas. I feel like for the first time that design is such an approachable concept and skill set to a lot more people.

2:28And it brings together sort of people who have aspirations for design and wanting to build things, wanting to prototype things, putting beautiful stuff out in the world much, much easier and faster. So Rio has gone through a lot of thinking and journey of, you know, what design means in the sense of having Cursor being part of the building blocks of it. I just really wanted to have him on the podcast and talk about the future of both coding and design. Ryo, you've been at Notion, now obviously ahead of Design Cursor. Why don't you take us through your journey and how you've been thinking about some of these topics?

3:06I think to me, it's like building software. there's like so many layers of abstractions and depth that you need to take care of. And in order to do something really great, you actually need to know everything or like you assemble a team that works really well together with people with different strengths on every layer. Maybe you have the greatest infra engineers, people doing ML. maybe you have you know really good design engineers who really like they can just hand write css and then they'll be like perfect um in order for say one person to do all of this or learn all of this it takes a long time of trial and error you have to build from the simplest things to you know gradually more complex scale it up to more people um share your workout to the public see what happens do this feedback loop and if you're doing it in a team sometimes it takes even longer because say you're just a designer you you're doing some mocks in figma and then you shared it out you got some feedback then your pm needs to do this like prd thing and then run more meetings and then there's like more people involved and then they're like and takes a long time.

4:30And then maybe like a year later, your design came out, but then it's like 20 % of what you wanted. But with this new thing, so with cursor, you can just say you have an idea. It might be a little ambiguous. You just tell it to tell the agent. And then it might give you something maybe like 60%, 70 % on the first shot. But you kind of skipped a lot of the complexities. We kind of transformed from like you need to understand all this thing, all these concepts of making software. Then you can do something too. I can do something now. and then I might get something that's maybe imperfect, not exactly what I want.

5:27But the process of iterating and kind of poking at it becomes really quick. And then as the agents evolve, as the models get better, as it talks to more tools, it understands visuals better. It can talk to Figma, the mocks that I had. it can talk to Notion, all the ideas, the docs, the meeting notes, anything. And the most important thing is it knows the code base, which is the truth, the material of how we build software. So that kind of changes the whole thing. It's like the tool not only impacts the individual software engineers, like for them maybe like for cursors like we kind of try to fit as many workflows and people as we can say there's like people who they pride themselves at like you know really think thoroughly write the most clean code for those people they can just type and then we do the tab thing and the tab thing got really good at like it's almost like it knows what you want to do next um so for those people they can do that but there's like increasingly more people doing the agents where like even for the most professional coders, they start to do those new things.

6:54Yeah. Yeah. I am trying to think of even myself as an example. Prior to joining the firm, I was on the product side. I was working with a lot of designers, design engineering teams. Back then it wasn't called that way. It's more like designers, front-end engineers. and then UX designers too. It was still a very disjointed workflow. It's like, you know, a lot of the design work happens more in isolated fashion with just the designers themselves. Like they spent two weeks coming up with a concept. What does the UI look like and work with UX designers on what the UX look like and then hand it off to the product team and works with engineers.

7:38Figma already bring that sort of process a lot closer that you can collaborate around one artifact that everybody can give in plus and prototype and bring sort of more of a close to the reality artifacts into these people's hands. where with Cursor, it's even one step closer is you can actually poke around and play with these artifacts that are functional and working. I'm curious, what does it mean for collaboration among these teams as you're mentioning the collapse of that iteration and the speed? And also, what does it mean for the different roles that's involved in design? Yes. So I think maybe like over the last, I don't know, 15 years or so, I think the art of making software fragmented a lot.

8:39And then we kind of split into different roles. Each role kind of used their own tool, used their own artifact. They think in their own kind of words and lingo. Say the designers are stuck in Figma before maybe it's like sketch. There were actually like, you know, files. and then the PMs, maybe they're like just writing docs and writing meetings or maybe they're in Google Docs and say the data people, maybe like, you know, some other tool and then everyone's like kind of siloed in their own way and then we need to kind of come up with a process to tie everything or like build better tools to kind of unify everything.

9:18We tried that at Notion but the problem is people have already developed so much habits. There's inertia to kind of change that or change people's tools or you can't really force anyone to adopt something new that doesn't perfectly fit them. But with AI, with cursor, things kind of flip again because we want to kind of build something where it can kind of connect and absorb all of these artifacts and formats. And later, maybe even within cursor, there could be different views of the same code. like showing the code as raw code or like as diffs is almost like just the very beginning. The act of making software is really just modifying the code.

10:32Like in some sense, like the PM writing the PRD is modifying the code, but they're doing it through a more passive, like organizational way. Maybe the designers influence more individuals. but when you do this you know disjointedly there's so much like back and forth collaboration issues as you grow the team it gets even more complex people start like breaking the software apart things are no longer simple no longer unified like we always talk about like you kind of ship your work chart type of thing and then the different roles kind of fight like designers are right the engineers are right the PMs are right but like you know there's this shared truth which is the code where like you know you can also gather a lot of information around putting everything synthesizing everything together then the agent can kind of handle all these things that you might actually not know fully but it kind of you know knows the truth it could know the presence which is maybe like Like, you know, what's in your code base?

11:45What are the actual, you know, running tasks or projects? Even gathering feedback or information from the real world. It could be also like, you know, from the past, say like all the knowledge that you've accumulated, your team preferences, like how the code base evolved in Git maybe. There's also the future, which is like, say you're planning ahead, you're like thinking about the vision, you're maybe like ideating some like big ideas you can actually do all of this with just one tool one agent but in mind you know for each individual user or team it might take a different shape and then like what we want is like there's like a base experience that almost works for anyone anything but you can get more specific if you know what you're doing or if you have specific needs you can even maybe you know use cursor as if you were using figma at some point in the future but the difference is you are not interacting with these siloed apps with their own like formats and then you don't have to do the conversion manually with meetings or whatever like it just does it then all you need to do is you're kind of thinking about the idea you're like iterating on it in whatever way that is the best for you.

13:14For the engineer, it might be like a code editor, but maybe for a designer, it's something more visual. For the PM, it might be something more like a document. That makes sense. I'm curious, given that there's a lot more focus and emphasis on this concept of taste after AI comes about, and now there's also this coworker that's an agent. helping with writing code, designing elements of the product. Where would the taste live? Where does that come from? Can you rely on agents for having good taste? Is that still heavily reliant on the creator as the human designer or developer? I don't really like people talking about taste as a word.

14:01Why is that? Because I think it's so ambiguous. Like how I see it as more like, I think taste is kind of like there's a part of like you're selecting out of you know all the options but in order to do that you have to kind of see everything or like you have seen it you have maybe dug into the past you have kind of figured out oh these are the ways people did this kind of thing you made a connection of you know some stuff from the past where you've seen in nature where, I don't know, some human made it or nature made it. And then you kind of connect that to your thing or the thing you want to do.

14:51Or you kind of, over time, develop like a preference. It's almost like you're self-selecting a boundary of this is what is right. This is what is beautiful. This is what is good. And I think it's very different for each person. There is no right. There is no wrong. It's more dependent on, say, like the things that you've seen. It's almost like an LLM. But the problem with an LLM is like it actually has seen everything. And it doesn't really have an opinion. or like it kind of confused itself thinking that people prefer purple gradients everywhere. But what is good is like the element because it has seen a lot of things, it can do the baseline really fast and really good.

15:50Then the thing on top is taste or like your self-selection of what is good. Like you're kind of drawing the boundary. That is your decision. Though the AI can increasingly help you do that. Say we have a new thing in cursor called plan mode. If you type in the prompt and then you don't really want to fill in the details, you can switch to plan, go. It will kind of just build the spec for you. And then you can add details. You can change it as you want.

16:26but it's almost like I don't really believe in say you give the agent something long running or like it's really like occluded like a really non-specific prompt and then expect it to give you exactly what you want like it's just not gonna work there needs to be something for the human to specify what is good what is right how I want to do it. If you don't put in that opinion, it will just produce AI slop. Yeah. You were alluding earlier to this sort of, you know, fight between product managers, designers, and engineers to some, or this sort of dance. And I know if you were saying a few years ago, the categories were somewhat different.

17:11As there is this blurring, how do you think these categories will evolve over the years?

17:22I think people always have their strength or their unique special skill or some spike. Say some people are more good at coordinating. Some people are good at the visual space. Some people are maybe good at architecting the lower level constructs. But I think of all of these people as just like they're software builders or makers. Like we kind of started there. Like if you look back when, so like the early computing era, there was no title or people who were maybe like researchers were like, but they maybe designed a low-level architecture. They maybe even, you know, built the UI and how to display the UI on the screen and the whole thing.

18:18Maybe one person or two or five. and when you did that, and also I think back then there were like less, say, economic constraints where they were funded and they weren't like trying to make money as much. So they kind of made the whole thing really whole. And now it's like you kind of break everything down. You try to optimize them with like processes and cost optimizations. and people become boxed into little areas and problem sets when the whole thing is actually all together. And that causes a lot of problems, I think. People now make software that, I don't know, like they don't even think of like like there's some like ideal that's kind of lost and people think too much on the technical problems, the design problems, the money problems, not the whole thing or what we're trying to make better for people.

19:35But, you know, we're kind of going backwards now as, you know, but they say tools like cursor. If you were, you know, self-identified as a designer or developer or something. Like I used to also struggle with this. I started like making stuff myself, the whole thing. And then I came to the US, I got a job titled product designer. I stopped coding. I made mocks and prototypes and I waited for them to happen for like years and they don't happen or like they ended up shipping as like a YouTube video. That's crazy. But, you know, with this new tool, a designer can build. They can actually like, you know, work on their craft, the stuff that they really care about, make that really good and let the rest, you know, be handled by the agent.

20:36they can you know kind of put their taste on top and all the stuff that they don't want to worry about give it to the agents but you can also assemble like a really good team so like there's like really good infra engineers front engineers PMs who are like not just running meetings get all of them together give them the same tool, the same code base. They can start covering each other's weaknesses and then amplify their strength. And the agent kind of holds everything. And, you know, instead of you pinging this guy, where is your design? It knows. Yeah, that resolves a lot of the common conflicts of spending more effort on the functional part of the software or spending more time on the artistic aesthetic side of the product itself and being appealing to the users.

21:41And you have worked at many of the very design-focused, design-centric companies from Notion to even prior to that Asana. Now, given sort of the democratization of who can touch that external-facing aesthetic part of the product, how do you think about influencing even, I guess, your current team and people at Cursor to spend more effort in thinking about that versus just the functional part of the product itself? Yeah. Yeah, one thing I want to call out is design is not just about aesthetics. It is actually, like, how I think of it is, like, it's all, it actually includes all the, say, the architectural designs or, like, all the concepts of what this thing is or like the company even.

22:34Say for Notion as an example, Notion is a pure conceptual product, meaning every single concept was designed by a person. So like in Notion, it is really just blocks, pages, databases, the workspace, and then everything kind of works around these concepts and then at every layer there's like a representation of them. There could be like, you know, at the very top is like the UI or like the brand or like the visuals, the aesthetics. But then there's actually the aesthetics, every layer, how you architect, say, like the front end code and architecture, how they, you know, how the reactive states sync and how you render stuff to like how do you like store these objects, how they, you know, relate to each other to all the core concepts of the thing.

23:30And if you look at software, it's actually really simple if you look at the concepts themselves. So design is kind of like trying to figure out what is the best configuration and the simplest state for all of us. Some people maybe only focus on the visuals or the interactions or certain slices. But I think the beauty is actually putting things all together as well as you can. So I think it is really about what I just talked about. It's like not seeing design as just, should we use a six pixel border radius or four? But it's rather like, how do I design the most simple system, fewest number of concepts, fewest code paths to do the most things for most people.

24:29You guys obviously have incredible product market fit with developers. We've alluded to a bunch of it, but can you share more about maybe either how you guys have navigated the idea maze of how you want to serve designers or just more around what kind of tooling you think there's an opportunity to provide? Yeah. So I think

24:52cursor is still our primary focus is on professional developers and teams. But because of that, people around them are already here. And I think for the longest time, we've been actually intentionally making cursor pretty hard to get in for, say, the non-technical people. But they're here now. And they actually struggle to get in. And they really want to get in.

25:28One example is like, when you open up cursor, there's like three buttons. It says open project, connect to SSH, clone repo or something. As a beginner or like a non-technical person, I can't understand any of this. But what if, say like, We just kind of give you the agent view blank. You can just start doing things. There's a lot of little things we can kind of fix to just make cursor feel more friendly and welcoming for these people who are kind of edge adjacent people who are maybe they know software concepts or certain layers, but they might not be able to code. I want to make sure that when they come in, they can, without feeling overwhelmed or feeling like, ah, this is like a code editor, it's an IDE, it is more like I can start doing things.

26:29And as I start doing things, I can maybe pick the paths that I prefer. Say like a designer, maybe they're just kind of chatting with the browser next to it. As the agent is making edits, they can kind of preview the changes. They can maybe interact in the browser, pick this element, change. Like, ah, I want to swap this with something else. And then boom, it happens.

27:01So how we do it is not, say, creating new products or splitting cursor. It is the same thing. But just different pre-configurations and packaging of the same thing. Because kind of like what I just said, thinking about the concepts, cursor itself is actually really simple. Or like AI agents in general are pretty simple. What you want to do is actually not... Like if you look at, I don't know, like a chat GPT agent versus like a cursor versus like a replant, a V0, a Notion agent even, the architecture or how they work, they're almost the same.

27:53So what if we can come up with a set of universal shared concepts for interacting with AI, with agents, with code, with software, but you can kind of mutate each one to fit more people. And to fit more use cases. And then each of them can leverage the best model to do this UI thing with the best view that fits me. I can configure it however I want. If I want to see everything, I can. If I don't want to see anything, I can too. This leads to my question of over the last few years, and I don't know if it's a few years or a decade, there is the concept of purpose-built tools for certain persona. whether it's Webflow for, you know, personal use cases like for landing pages, for cell v0 for more of the, you know, front-end developers building Next.js apps.

28:56There's tools for designers. There are tools for handoff from design to engineering, where now there's more of a concept of the everything app. ChatGPT is kind of everything app. Notion is kind of everything app. You can go to it for your note-taking, but you can also publish in Notion Pages. Curse is becoming more of an everything app. Is that a path that we're going down towards of having these all-encompassing apps that can do a lot more things that used to be captured by a single-purpose app? Is there still a place for purpose-built tools for a specific use case or persona? How do you see that dynamic?

29:33I think it's just like different philosophies of doing things and making software. I think there's almost two ways you can look at the thing. There's the user-centric, human-centered design path, which is you start from a problem. You identify the group of people who have this problem, figure out what they want, build really specific solutions for them. versus like there's the more system angle to think about things where you're just kind of looking at the software itself, how it is composed, and then think about, where do I tweak a little bit to satisfy this constraint or to make this use case work or to enable this tool to work for these people?

30:33I think it's like fundamentally two different philosophies I think it is much easier to do the user-centric way,

30:47but it kind of limits you from the beginning because when you start building these specific solutions, they only work for those specific people. If you want to grow the people or you want to grow the use cases, you actually need to kind of tear apart everything you have, your core concepts. And a lot of people just can't do that. So what they do is instead of doing that, they add more things, more concepts, more features. And then there will be a point where this thing no longer serves your initial group of people anymore. The simple thing is no longer simple. The purposeful thing is no longer simple.

31:30And all of these purposeful apps, they're kind of selfish. They are siloing people, siloing workflows, file formats, creating islands. If you look at the thing, like all these purposeful apps, whatever. I also work at Asana. Asana, the core concepts are really tasks and projects. Everything around tasks and projects. Everything they add needs to work with those. And that naturally limits what it can do. versus say like Notion. Like how we see Notion, it is not note-taking. It is disguised as note-taking. Like you come in, you can start from a blank page, you can type, but then what you're doing is actually like, you know, blog pages, databases, and the workspace.

32:26Each block is almost like a JSON object. A page is just an array of JSON objects, and then we render each block in the, you know, the layout and the type it is. And then you can put them in a database. Now they have, you know, more properties. They share more stuff. There's more hierarchy. And all pages can nest each other. That is Notion. But then you can do whatever with it. You can have a task project database. They all work together. They can be a list. They can be a board. Do whatever you want. But then the problem is, you know, for these more universal type of apps, It's like, because it's so open-ended, it's kind of hard to get started.

33:10If I don't have the patience to kind of figure out how it works, I might not even get to the tests and projects. So there's always that tension. But it is fixable. You can build better packaging. You can use AI. so I think there's just like my personal preferences I would try to build something that works better for everyone than just these people are the people we care about I don't care about everything else and then I think they should use my thing that's not how you do it we talk about AI We talk about agents and we talk about how it really speeds up building things and prototyping. But when it really comes to these type of, you know, helping users to understand a product better, onboarding, learning the new concepts, also to you as a designer, design leader, how does like interacting with AI really improves the usability utility of the products?