When AI starts writing code, will the traditional interface disappear? Ryo Lu, head of design at Cursor, gave the answer: it’s not that the interface is gone, but that the “interface” will be redefined as a mapping of everyone’s mind. This article exclusively shows how he merged five fragmented functions into one AI Agent, allowing designers to write 130,000 lines of code directly, and one person to “atmosphere” a complete operating system in two months, and predicted that “over-specialization” will fail in the next ten years – the real competitiveness is the systematic thinking of “talking to AI”.
Have you ever wondered that traditional software interface design might be coming to an end? When I listened to this interview with Ryo Lu, Cursor’s Head of Design, I realized that we were at a historic turning point. The interface elements we’re used to—buttons, menus, forms—may soon be replaced by a whole new way of interacting. What’s even more shocking is that Ryo makes a point that sounds like science fiction: the interface of the future will map directly to the way we think. This is not some distant fantasy, but a real transformation that is happening within Cursor.
As the head of design at one of the most influential companies in the tech world today, Ryo Lu’s experience is a legend in itself. He was an early designer at Notion and was involved in building the product that changed the way countless people work. But when he turned to Cursor, his challenge was even more complex: how to design an interface for an AI-powered programming tool that would serve both professional developers and make it accessible to “vibe coders” who didn’t know anything about programming. At the heart of this challenge lies that Cursor is more than just a code editor—it’s more like an intelligent programming companion that understands user intent and automates complex programming tasks.
From Five Concepts to One System: Cursor’s Path to Design Unity
One of the first things Ryo did when he joined Cursor was to solve a seemingly simple but actually crucial problem: conceptual unity. At that time, Cursor had five different features: Tab autocomplete, Command K inline editing, Chat conversations, Composer code generation, and Cursor Agent automation mode. These features are developed separately and have different names and shortcuts, and users often have trouble figuring out which one to use.
“What I do is say that all these things are the same thing. They are all agents. Then we combine them into a single concept. Ryo described his improvement as follows. It sounds simple, but the power of this systematic thinking is enormous. With this unification, Cursor’s user experience has taken a qualitative leap. Previously, users needed to learn five different interaction patterns, but now they only need to understand one core concept: AI agent. This simplification is not subtraction, but refinement – while retaining all functions while greatly reducing the cognitive burden on users.
To achieve these three challenges, product managers will only continue to appreciate
Good product managers are very scarce, and product managers who understand users, business, and data are still in demand when they go out of the Internet. On the contrary, if you only do simple communication, inefficient execution, and shallow thinking, I am afraid that you will not be able to go through the torrent of the next 3-5 years.
View details >
This reminds me of a timeless truth in software design: the best design is often not about adding more functionality, but about finding the right level of abstraction. Just like the iPhone was revolutionary not in adding more buttons, but in replacing all physical keys with a touchscreen. What Ryo does with Cursor is essentially the same: finding a higher level of abstraction that makes complex functionality simple and easy to understand.
I particularly appreciate one point Ryo mentioned: “We want to serve everyone, from the most experienced programmers who want full manual control, to the ‘atmosphere’ users who prefer to let the agent do everything. There are a lot of different small configurations, patterns, and behaviors in this spectrum. You want to serve everyone, but give everyone the ideal place on this spectrum. “This design philosophy is very profound—not to force all users to use the same way, but to create a system that is flexible enough for different types of users to find their own way of interacting.
The effect of this unification is remarkable. Ryo mentioned that many senior engineers had been using Cursor before, but had no idea that the Agent feature existed. After the unification of concepts, this problem disappeared and all users defaulted to Agent mode, which is probably one of the most important reasons for Cursor’s rapid growth since February of this year. This case perfectly illustrates how good design can unlock the true potential of a product.
Redefining the Role of Designers: From Interfaces to Systems
What struck me the most during my conversation with Ryo was his reflections on the designer’s role shift. The traditional design process is like this: a designer draws a beautiful interface in Figma, and then engineers build a product based on these static images. But in the age of AI, this way of working is becoming obsolete.
“The things we design as designers, they just go up a notch. Instead of designing exactly what this UI interface looks like, you’re actually designing a container. Ryo explains the new design paradigm this way. This concept of “container” is very important – designers no longer design fixed interfaces, but design a system that can adapt to different users and different situations.
The underlying reason for this shift is that AI interfaces are highly dynamic. Traditional interfaces are static, and every user sees the same buttons and menus. But the AI interface automatically adjusts based on the user’s behavior, preferences, and current task. In this case, the designer’s job becomes the rules and boundaries that define these dynamic changes.
Ryo illustrates this with a vivid metaphor: “The ideal interface is different for everyone. “This means that designers need to move from designing a single interface to designing interface systems. It’s like an architect doesn’t design a house, but a modular system that can be combined according to different needs.
I think this shift has far-reaching implications for the design industry as a whole. Designers need to have a stronger systematic mindset and understand how to build flexible and coherent design systems. At the same time, designers also need to have a deeper understanding of technology, as the boundaries between design and technology are blurring in the age of AI. As Ryo says, “If you know how to interact with AI, you’re almost leveling the field.” ”
This brings me to another important point: with AI assistance, over-specialization may no longer be an advantage. Ryo observed that some 17-year-olds “ambient programming” with AI create more amazing things than some big tech engineers writing 10 lines of code a day. This shows that in the age of AI, it’s not about how much expertise you have, but how you effectively collaborate with AI.
New Challenges in the AI Agent Era: From Interface to Conversation
When discussing the future development of Cursor, Ryo encountered a seemingly simple but actually very complex problem: how to manage multiple AI agents running in parallel. Imagine having five AI agents handle different programming tasks at once, and you need to be able to see their progress, manage their outputs, and decide what to do with their results.
“I’ve probably spent most of the month thinking about this. How do I start multiple agents and manage them, see what’s going on, and then what do I do with them once they’re done? Ryo described the complexity of this challenge. Users need to be able to plan tasks, execute tasks, review results, and then decide how to incorporate those changes.
Interestingly, after much thought, Ryo discovered that this complex problem essentially boils down to a very familiar concept: “Oh my God, this is the to-do list again.” We’re back here. Every time we go back to the to-do list. “This discovery is both frustrating and exciting. Frustrated because it looks like we’re reinventing the wheel, excited because it means we can take advantage of concepts that people are already familiar with.
But it’s not a simple to-do list. The key difference is that these tasks may be performed by AI agents rather than humans. This seemingly small difference actually makes a huge difference. AI agents can handle multiple tasks in parallel, can work over longer time spans, and can handle more complex tasks than humans. But at the same time, they also require more complex management and supervision mechanisms.
I think this example perfectly illustrates an important principle of AI product design: the best AI interfaces tend to wrap complex AI capabilities in concepts that people are already familiar with. Users don’t need to learn new interaction patterns, but they can gain capabilities that go far beyond traditional tools. This design philosophy can be seen throughout Cursor’s overall architecture.
Ryo also mentioned an important point: this task management system serves not only AI agents but also humans. “Imagine breaking down the entire software development process into a bunch of tasks that are associated with chat prompts. Because the concept is so generic, the interface doesn’t have to be complicated, and anyone who has seen the list view can start doing these things. “This unified task management system allows humans and AI to collaborate within the same framework.
From Code to Thinking: The Ultimate Evolution of Interface Design
At the end of the interview, Ryo shared a point that sounds like science fiction but is convinced that it will become a reality: “The interface will become like the way you think.” I think it will get closer to you. Maybe we won’t operate through a proxy device. Maybe it’s even closer. I am a visual thinker. I still see my canvases, but maybe they are suspended in my mind, not on the screen. ”
This prediction made me think about it for a long time. Now we interact with the computer by tapping, dragging, typing. But what if the interface maps directly to our mindset? For visual thinkers, information may be presented in the form of graphics and spatial relationships. For logical thinkers, interfaces may be organized more in the form of causality and decision trees. For emotional thinkers, interfaces may convey information through color, rhythm, and mood.
This personalization is not just a superficial customization, but a deep cognitive match. As Ryo says, “The ideal interface is different for everyone.” But this difference is not arbitrary, but based on each person’s unique way of thinking and cognitive patterns.
I think this evolution has begun to some extent. When I use AI tools like Cursor, I find myself thinking less and less about “how to operate this software” and more about “what do I want to achieve.” The software began to understand my intentions instead of asking me to learn its language. This shift, although still in its early stages, is already a clear trend.
Brain-computer interface technologies like Neuralink, which Ryo mentioned, make this future even more concrete. While it may be years before we see true CBI widespread, the concept itself is no longer a fantasy. When interfaces have direct access to our minds, the role of designers will change radically again.
I find it most interesting that even in that future, there will still be a need for someone to “design concepts and ideas, clarify them, reduce them to the simplest, unchanging form”. Technology may change the form of interfaces, but the core of design thinking – understanding user needs, simplifying complexity, creating intuitive experiences – will always be valuable.
The Power of Personal Projects: The Future of Creation from RyOS
In the interview, Ryo shared the story of his personal project, RyOS, which perfectly showcases the potential of individual creativity in the age of AI. RyOS is a complete operating system interface with multiple applications, window management, file system, and other complex features. Amazingly, this project of 130,000 lines of code was mainly done by Ryo alone using Cursor in one to two months.
“I asked ChatGPT how long it would take the average engineering team to build it. The answer is a few months or years, and it takes dozens or 20 people. And this is just atmosphere programming that I do alone in my spare time. “The contrast is shocking. Projects that traditionally took a team months or even years to complete can now be achieved in weeks by one person with AI tools.
What’s more, the origin of the RyOS project is entirely accidental. Ryo originally just wanted to make a soundboard app for the team leaving Notion so that his colleagues could hear him in meetings. But with the help of AI agents, this simple idea gradually evolved into a complete operating system. “No plan, purely feeling.” Ryo describes the whole process as follows.
I think this example illustrates an important feature created in the age of AI: the distance from idea to realization is shrinking dramatically. In the past, even if you had a good idea, you needed to assemble a team, assign tasks, and coordinate progress, and the whole process could take months. But now, you can start implementing it the moment you have an idea, allowing it to evolve naturally as it comes to life.
The impact of this change on innovation is far-reaching. When the cost of implementation is significantly reduced, we can experiment with more ideas and experiment more. Ideas that were previously abandoned because they were too costly are now viable. This means that the bottleneck of innovation is no longer the ability to achieve, but imagination and judgment.
One point Ryo mentions is particularly thought-provoking: “One of the reasons I joined Cursor was that I wanted the gap from having an idea to becoming a reality was close to zero. “It’s not just about improving the efficiency of tools, it’s about changing the way we think. When you know that any idea can be implemented quickly, you will be more daring to think about things that seem impossible.
This also explains why Ryo was able to evolve from a soundboard application to a full operating system so quickly. In the traditional development model, each additional feature requires careful consideration of cost and time. But in AI-assisted development, trial and error costs become extremely low, and you can always try new directions and see what happens.
Fundamental changes in team structure
Ryo’s experience also made me think about the impact of AI on team structure. Traditional software development requires a clear division of roles: product managers are responsible for requirements, designers are responsible for interfaces, front-end engineers are responsible for implementation, back-end engineers are responsible for logic, and test engineers are responsible for quality. Each role has clear boundaries and handoff points.
But in AI-assisted development, these boundaries are blurring. Ryo, as a designer, can now write product code directly. He doesn’t need to wait for engineers to implement his design, nor does he need to communicate his ideas through detailed documentation. He can directly create feature prototypes, allowing the team to experience realistic interactions.
“The engineers see us more as part of them.” Ryo describes the change as follows. I think the value of this change is not only in improving efficiency, but also in reducing communication loss. In the traditional design-to-development process, some details are lost every time the message is passed. The designer’s intentions are communicated to engineers through documentation and meetings, but engineers may have different understandings. This multiple passes result in a significant difference in the final product from the original design.
When designers can write code directly, this information loss is greatly reduced. Designers can express their intentions precisely without relying on others’ interpretations. At the same time, they can better understand the technical constraints and consider the feasibility of implementation during the design phase.
This change has also impacted recruitment strategies. As mentioned in the interview: “We have two designers, which means we have two front-end engineers. When you don’t need someone to be very proficient in front-end development, the type of engineer you need to hire also changes because I can do whatever function we are developing. ”
I think this indicates an important trend in the future of team structure: less specialization and more important cross-disciplinary capabilities. Team members no longer need to achieve the ultimate level of expertise in a narrow field, but need to have the ability to quickly learn and apply in multiple fields. AI tools are lowering the barrier to entry in various professional fields, allowing more people to participate in all aspects of product creation.
The timeless value of design thinking
Despite the rapid development of technology, I don’t think the core value of design thinking will disappear with the advent of AI. As Ryo says, “At the code level, we can do almost anything.” “But the question is not what we can do, but what we should do.
The value of designers in the AI era is mainly reflected in several aspects: the first is conceptual clarification. The more complex the technology, the more someone needs to be able to simplify complex concepts into a form that users can understand. Ryo’s unification of Cursor’s five functions into a single Agent concept is an exemplification of this capability.
The second is systems thinking. As product features become more and more complex, someone needs to be able to think holistically about how to organize these functions into a coherent system. This capability is not a replacement for AI, as it requires a deep understanding of user needs, business objectives, and technical constraints.
The third is aesthetic judgment. Even if AI can generate interfaces, what constitutes good design still requires human judgment. Aesthetics are not just visually appealing, but also a comprehensive embodiment of functionality, ease of use, and emotional experience. This judgment ability requires long-term training and accumulation, and is the core competitiveness of designers.
The fourth is empathy. The essence of design is to understand the needs and pain points of users and then create solutions. This empathy is unique to humans, where AI can analyze data and patterns but cannot truly understand human emotions and motivations.
I particularly agree with Ryo’s point: “The end result is beautiful, high-quality software.” “This goal will not change no matter how technology evolves. The tools we use may change, the way we work may change, but the fundamental goal of creating beautiful, useful products is eternal.
At the same time, I also think that designers need to adapt to new ways of working. They need to learn to collaborate with AI and understand its capabilities and limitations. They need to move from designing static interfaces to designing dynamic systems. They need to move from communicating ideas to directly realizing them. These are new skills, but they are built on traditional design thinking.
Rethink professional boundaries
Ryo’s experience has left me pondering the larger question: What is the significance of professional boundaries in the age of AI? Traditionally, we have broken down complex work into different areas of expertise, each deeply specialized in their own field. But is this division of labor necessary when AI can compensate for our lack of expertise?
“I don’t think over-specialization makes sense at all in the next 10 years or so.” Ryo’s point of view is challenging. He does not mean that professional skills are not important, but that deep expertise in a single area may not be as important as the ability to integrate across fields.
I’ve seen many examples of young people who quickly mastered multiple skills through AI tools to create amazing work. They may not be experts in any one field, but they know how to combine different skills to solve problems. In contrast, some people who have received highly specialized training under the traditional framework have difficulty adapting to new ways of working due to the solidification of their thinking.
This reminds me of the polymaths of the Renaissance. Leonardo da Vinci was both an artist, an engineer, a scientist and an inventor. The boundaries of knowledge in that era were not as clear as they are now, and it was easier for people to move freely between multiple fields. Perhaps the AI era will bring us back to a similar state, where technology lowers the barrier to expertise and allows more people to become creators across disciplines.
But I also think that does not mean that professional skills will disappear completely. Instead, I feel like there’s a new trend of specialization: specialization in collaborating with AI. Just as Ryo specializes in design and development with AI, various specialized roles that work with AI may emerge in the future. The core of these roles is not mastering a traditional skill but knowing how to work effectively with AI for complex tasks.
A fundamental shift in design philosophy: from universality to individuality
Ryo’s idea that “the ideal interface is different for everyone” made me realize that we are going through one of the most important philosophical shifts in the history of design: from “designing a perfect interface for everyone” to “designing the best interface for everyone”. The implications of this shift are far more profound than we think.
Traditional design philosophy is based on the assumption that there is an “optimal solution”, a universal interface that meets the needs of most users. This way of thinking is deeply rooted in the logic of mass production in the industrial era. Just like Ford’s “any color is fine, as long as it’s black”, we’re always looking for a standardized interface that can serve everyone. This approach is indeed effective in reducing costs and improving efficiency, but it also poses a fundamental problem: in order to care for most people, we often ignore the unique needs of individuals.
I’ve seen countless products fail because of this “one-size-fits-all” approach to design. Designers spend a lot of time researching the “average user” trying to find the perfect interface that balances all needs. But the reality is that the average user does not exist. Each real user has their own unique usage habits, cognitive patterns, and preferences. When we design for a fictional average user, we are actually designing for someone who doesn’t exist.
Ryo’s work at Cursor opened my eyes to another possibility. “We want to serve everyone, from the most experienced programmers who want full manual control, to the ‘atmospheric’ users who prefer to let the agent do everything,” he mentioned. “It’s not about trying to find a middle ground to balance these very different needs, but about building a system that can adapt to different needs.
At the heart of this transformation lies the improvement of technical capabilities. Previously, customizing the interface for each user was not technically feasible and costly. But AI technology makes this personalization possible. The AI can understand the user’s behavior patterns, preferences, and current tasks in real-time, and then dynamically adjust the interface to match the user’s needs. It’s not a simple customization, but a deep cognitive match.
I think this shift will have three levels of impact. The first is the impact of interaction. Traditional interface design strives for consistency, with all buttons in the same place and all menus in the same structure. But in the age of personalization, the definition of consistency needs to be reconsidered. Perhaps the most consistent experience for an experienced user is to give them quick access to advanced features; And for newbies, the most consistent experience is the simplified interface and step-by-step guidance.
The second is the impact of cognition. Ryo mentions being a visual thinker and that he wants to see a canvas suspended in his mind. This made me think, what if the interface could match the user’s cognitive patterns? Visual thinkers may be better suited for graphical information presentations, logical thinkers may prefer layered lists, and emotional thinkers may need more color and motion feedback. When the interface can adapt to these different cognitive patterns, user interaction with the software becomes more natural and efficient.
The third is the philosophical influence. This shift is actually challenging the fundamental assumptions of design. Instead of assuming that there is a “right” design, we believe that the value of design lies in its ability to adapt and serve diversity. This way of thinking affects not only product design, but also our understanding of users, technology, and innovation.
But I also see the challenges that come with this shift. The first is complexity management. When each user sees a different interface, how to ensure product integrity and brand consistency? How can support and documentation be provided for such a system? How to test and optimize a kaleidoscopic interface?
The second is the transformation of designer skills. Traditional designers need to master skills such as visual design and interaction design. But in the era of personalization, designers are more like system architects, and they need to design not specific interfaces, but the rules and algorithms generated by the interface. This requires stronger systems thinking, data analysis capabilities, and even programming skills.
There is also a deeper question: when the interface is fully personalized, do we lose a common design language? Traditionally, standardized interfaces have made it easy for users to migrate between different products. But if the interface is different for each product and every user, will this migration cost increase significantly?
I think the answer lies in finding a balance between personalization and standardization. As Ryo said, it is necessary to identify those “underlying basic units that will never change.” Some concepts and interaction modes are universal, such as basic operations such as clicking, dragging, scrolling, etc. Personalization should be built on these stabilities, not completely reinventing interactive language.
From a business perspective, this shift also brings new opportunities and challenges. Products with higher personalization tend to be more sticky because migration costs are higher. But at the same time, developing and maintaining such a product also requires more resources and technical investment. This could exacerbate the gap between technology companies, and those that can master personalization will gain a significant advantage.
I believe we are in the early stages of this transformation. Nowadays, personalization is mainly superficial, such as theme colors, layout options, etc. But as AI technology evolves, we will see a deeper level of personalization, and the interface will be able to understand and adapt to the user’s thought patterns, work habits, and even emotional state. This will no longer be the personalization of the interface, but the personalization of the experience.
Deep thinking about the future
Through Ryo’s sharing, I saw the possible future direction of software development and product design. This is not just an upgrade of tools, but a fundamental change in the entire creative process. We are moving from an industrialization model based on the division of labor to an intelligent model based on collaboration.
In the traditional model, creating a product requires a strict plan, a clear division of labor, and complex coordination. Each step has clear inputs and outputs, and each role has fixed responsibilities. This model is effective when dealing with complex projects, but it also comes with significant coordination costs and communication losses.
In the new model, the creative process becomes more like improvisation. You have a general direction and then adjust and improve as you implement it. AI tools make this way of working possible because they greatly reduce trial and error costs. You can quickly experiment with different ideas to see what works and what doesn’t.
This change has far-reaching implications for individuals and organizations alike. For individuals, it means the need to develop greater adaptability and learning. You need to be able to quickly master new tools, collaborate effectively with AI, and make decisions in uncertainty. For organizations, it means the need to build more flexible structures and cultures that encourage experimentation and innovation, and tolerate failure and adjustment.
I also thought about the challenges that this change could bring. When the cost of creation is significantly reduced, more products and services will appear on the market. This means that the competition will be fiercer and the user’s attention will be more distracted. In this environment, what really matters may not be what you can create, but what you choose to create. Judgment and taste may become more important than technical ability.
Another challenge is quality control. When anyone can create complex products quickly, how can you ensure the quality and safety of those products? This may require new governance mechanisms and standards systems. We need to find a balance between encouraging innovation and ensuring quality.
Finally, I think the biggest inspiration of Ryo’s story is that the future belongs to those who dare to experiment, who are good at learning, and who can adapt to change. No matter where you are now, no matter what your background is, as long as you keep an open mind and a passion for learning, you can find your place in this changing era. The key is to take action, start experimenting, and start creating. As Ryo said, “The gap from having an idea to becoming a reality is approaching zero.” “Now is the best time to start.
