“The AI hardware market does not have a correct answer; deeply focusing on one product can easily lead to misconceptions. ”
Author丨Wu You
Editor丨Liu Wei
Entering the office of Li Yunzhao, CEO of XuanYuan Technology, the desk is filled with figurines from “Nezha”, including Nezha, the Boundary Beast, and Shiji Niangniang… Li admires Nezha’s spirit of breaking through and resilience. These figurines are not just silent collectibles to him; they are anchors that provide emotional value, a sense of ownership, and satisfaction.
In his view, even without AI figurines, emotional value and companionship can still be provided. Therefore, he firmly asserts that “AI companionship” should not be regarded as an independent industry, but rather as a derivative effect and result. Li emphasizes that XuanYuan Technology is essentially an AI robotics technology company, not an AI companion toy enterprise.
As a PhD from Harbin Institute of Technology’s Robotics Institute and a serial entrepreneur, Li led his team to win China’s first international championship in heavyweight combat robotics during his university years. His last venture focused on educational robots, which took six years to become profitable, but he also paid a nine-digit “tuition fee”. Now, riding the AI wave, he founded XuanYuan Technology, aiming to create consumer-grade AI robots that accompany users throughout their entire life cycle.
Li is very confident in the company’s technical capabilities, especially in the fields of continuous perception memory and multimodal information fusion, hoping to give hardware more “life-like qualities”. On-site, he had the Yonbo robot introduce the company, and Yonbo responded in the voice of Zhu Bajie from “Black Myth: Wukong”: “X-Origin specializes in various disputes… emotional recognition, personality evolution, full-link closed loop, which is not something those shell AI can compare to.” Li jokingly called this Yonbo his “best partner” on his entrepreneurial journey.
Despite his technical confidence, Li remains humble, clearly stating that he is responsible for technology, strategy, and investment. When discussing product details, he actively invites CPO Wei Yingda to join the conversation, providing a more professional perspective.
01
Starting a Business with 8 Partners for 7 Years: The Team is Back-to-Back Logic
Lei Feng Network·Whale Rhino: You started your entrepreneurial journey in undergraduate studies, received special training from Harbin Institute of Technology, led your team to win China’s first heavyweight combat robotics international championship, later incubated ToB XuanChuang Robotics, and now founded XuanYuan to create AI companion hardware. What are the differences between these two entrepreneurial experiences?
Li Yunzhao:Entrepreneurship can be divided into three stages: the first stage is paying tuition, the second stage is doing business, and the third stage is building a career. My first entrepreneurship was in the tuition-paying stage, where I was cultivated as an innovative and entrepreneurial talent in school, focusing more on practice, learning, and understanding the essence of entrepreneurship. Now, XuanYuan is in the practical implementation stage, where we have sorted out a complete business model and established strategic positioning.
Why the transition from XuanZhi to XuanYuan? Because we believe AI is an opportunity to reshape history. Since we want to do a grounded business, we need to think about the core demands of the AI era: what are the fundamental needs of humanity? AI is not just an accumulation of tools, but a reconstruction of needs.
Lei Feng Network·Whale Rhino: How much tuition did you pay in your last entrepreneurship? What were the main areas?
Li Yunzhao:At least nine digits. The highest proportion of tuition was in product supply chain and business lines. No product is simple; behind the seemingly easy surface are countless details and traps. Our first-generation product had a definition error, the supply chain collapsed, costing us 30 million in tuition; entering the education sector had a low ceiling, and ToB/ToG business receivables were difficult, with enormous pressure during special periods.
Entrepreneurship is like a foggy forest, with chaotic flowers that can confuse the eyes. If the main strategy is unclear and one tries to do everything, nothing will be accomplished. However, these tuition fees have transformed into XuanYuan’s advantages: a team that has fought hard battles and a mature supply chain, which is an invisible moat for entrepreneurship.
However, we have not paid much tuition in terms of hiring. I have been a leader among children since I was young, and I have always had a smooth relationship with the team, where everyone can find their position and footing, and all the “gods” are in place.
Lei Feng Network·Whale Rhino: You have 8 partners who have been with you for 7 years, which is uncommon in the entrepreneurial circle. What qualities or entrepreneurial philosophies do you think have allowed the team to stay back-to-back for so long?
Li Yunzhao:Initially driven by passion, the key in the later stages is to clarify the matters, distribute what needs to be shared, and give the team expectations and vision.
Warren Buffett’s last speech left me with three memorable points: find a worthy cause to fight for, find companions, and lifelong learning.
My personal abilities may be somewhat informal, and I do not pay much attention to many details; the key is to utilize people’s strengths and let everyone shine. After forming the current team, I have never envied any other team; our team is the best. If we cannot succeed, others will find it even harder.
Wei Yingda:I can add to this question. I was originally a member of Li’s rival team. We were all in Harbin at the time, and Li’s persistent spirit made me feel he would definitely succeed. The first time I met him, he was developing in a very small office, working on the table and resting under it, very frugal. Even now, he spends most of his time in the company.
Li invited me to join in 2017 and kept inviting me until I agreed in 2021.
02
There Are Only Two and a Half Directions for AI to Land: AI Companionship is a Pseudo-Demand
Lei Feng Network·Whale Rhino: Over the past 7 years, transitioning from combat robots to companion robots, have you considered what technology can and cannot do during this process?
Li Yunzhao:I believe there are only two and a half directions for AI technology to land: the first is the cost-reducing and efficiency-increasing tool attribute, which for individuals means practicing spoken language, writing articles, translating, etc., and for enterprises, it means cost reduction and efficiency increase from another dimension; the second is providing emotional value, understanding what you say, and giving better feedback; the third is the “magical moment” of cross-era interaction, which can be understood as products with better interaction quality than before, such as AI glasses, but currently, the industry’s technological development cannot meet human expectations for AI glasses, so it can only be considered half a direction.
XuanYuan’s current work leans more towards tool attributes and providing emotional value, following the path of consumer-grade AI products.
Wei Yingda:From a product manager’s perspective, I do not get too hung up on technological boundaries, but rather focus on technology combinations, utilizing existing technologies to see what can be pieced together. Mobile phones and new energy vehicles are essentially doing combinations. The arrival of this AI era has brought a significant qualitative change in AI’s pure text interaction capabilities.
Lei Feng Network·Whale Rhino: In fact, you have been thinking about AI companionship for a long time.
Li Yunzhao:I think more about the development direction of the AI era. In XuanYuan Technology’s vision, there is not a single mention of companionship because AI companionship is a pseudo-demand and a false proposition that cannot become an industry.
Our human friends not only have companionship functions but also more tool attributes, such as helping with moving, fetching water, and can do many things together. The companionship effect ultimately presents itself through daily interactions. Therefore, there should first be a biological entity that resembles a friend to create companionship; companionship itself is passive.
In other words, when technology cannot reach the level of a human, true companionship cannot be achieved. I just mentioned that the provision of emotional value is closely related to companionship, but providing emotional value does not equate to companionship.
Wei Yingda:We often joke internally that AI companionship is a pseudo-demand. If a person has a high spiritual pursuit, they do not need AI to pass the time. The optimal solution for human companionship is to have many friends, many close friends, and many confidants; AI companionship is not the optimal solution, but it can fill the psychological gap when there are no other human companions, providing an outlet for expression.
Now everyone likes to pretend to be strong in public, hiding their weak side. If information security is done well enough, people can also reveal their vulnerable side to AI robots, which is beneficial for mental health.
Lei Feng Network·Whale Rhino: Based on your understanding of companionship, what misconceptions do you think current players in the AI companion hardware market have regarding their products?
Li Yunzhao:Many products are purely riding on the concept of AI companionship, lacking understanding of AI technology. Existing technology is insufficient to recognize physical space and human emotions; so-called companionship is a dimensional reduction of emotional value. Ultimately, it is just a pile of functions or forced creation.
Wei Yingda:Companionship must generate emotional value, and there are many types and ways to do this. For example, our game development also provides a form of companionship. Adding AI transforms linear companionship into intelligent companionship, where the fundamental meaning of companionship is to encourage users to invest more time.
Robot companionship is a new biological form that humans have fantasized about for a long time and will inevitably exist, positioned between pets and humans, capable of understanding clear instructions and exhibiting loyalty. Current technology can achieve robot companionship, but at this stage, it resembles a child.
Lei Feng Network·Whale Rhino: How do you view voice-interactive AI companionship? There are already many software products that focus on companionship; why do you still want to create hardware?
Li Yunzhao:Voice is the best interaction method, but it needs to be supplemented with the ability to analyze other factors, such as being able to see your expressions and perceive the surrounding environment, helping to solve complex problems in complex environments. Starting with voice interaction in AI hardware is not a problem; the key is how to provide more accurate and intelligent feedback to people in this dimension, which is a question that everyone needs to research deeply.
Wei Yingda:Voice interaction is similar to the basic memory logic of humans, saving conversations as events. With the rapid development of large model technology as a foundation, it has been done well, but purely software-based dialogue products are easily abandoned because they lack a sense of ownership. An app can become a fraction of a percent, but independent hardware can remain online continuously. In the future, hardware will resemble human partners rather than mere toys.
03
Creating AI Hardware with a Sense of Life: Achieving Continuous Perception Memory
Lei Feng Network·Whale Rhino: XuanYuan’s vision is to create consumer-grade robots with a sense of life. How do you understand the “sense of life” in hardware?
Li Yunzhao:We aim to create hardware products with a sense of life by achieving continuous perception memory, emotional drive, and self-maintenance capabilities, building a true artificial life system. This is also the ultimate goal of AGI.
The core difference between current AI and AGI lies in analytical and reasoning capabilities. To accurately assess scenarios and situations and provide appropriate feedback, multimodal information fusion must be achieved. Current multimodal fusion faces significant challenges, and visual bionic systems rely on real-time perception. We have successfully broken through these bottlenecks, enhancing decision-making accuracy. TTS technology achieves near-zero latency, and voiceprints can be replicated in just three seconds. We are also tackling complex scene multi-person dialogue technology.
Wei Yingda:The sense of life that Li is responsible for and the sense of life that I am responsible for are somewhat different. Li’s pursuit of a sense of life is truly natural feedback, with responses that have spirit and richer voice. My pursuit of a sense of life involves some “shortcuts” in the productization process to enhance user experience.
For example, adding presets to deepen conversations; giving the robot expressions when turning on and off. We will also reference game development methods, adding story backgrounds to AI characters, making their personas more complex, stimulating users’ desire to communicate. Games are products that easily create dependency and make people lose track of time; replicating games into robot entities completes half of the task.
Lei Feng Network·Whale Rhino: The first Yonbo targets children; what considerations led to this decision?
Wei Yingda:Actually, this is my personal wish; the initial intention of creating Yonbo was to give my son a romantic gift for Children’s Day. My son is 7 years old, and during the product development process, we found that there are very few meaningful products available for children aged 4-8. Traditional toys like learning machines and writing pads do not cater to children’s desire for expression; almost all toys suppress children’s expression. We also hope that children can stay a little further away from mobile phones and other indoctrination-type products, which is the core motivation for creating this robot.
Lei Feng Network·Whale Rhino: Children often get bored quickly. The standard version of Yonbo is priced at $799; how do you persuade parents to pay?
Wei Yingda:It really comes down to how to reduce the product’s rate of being unused. The products with the lowest rate of being unused in the market are mobile phones, computers, and game consoles. Why are these products less likely to be unused? Because their content is continuously updated. Yonbo has something like a game manual; we will launch a major game task every 2-3 months, allowing children to experience new tasks. Yonbo remains the same, but the games are updated.
We will also create a full-scene consumer product based on personal focus AI models, hoping that after users use one XuanYuan product, they can generate a personal XuanYuan memory. At that point, when we launch other products, memory can be seamlessly transferred, similar to changing to a new iPhone, making the transition less painful.
Lei Feng Network·Whale Rhino: Have you considered emulating the logic of Xiao Tian Cai by adding social features to the product to increase user stickiness?
Wei Yingda:For a newly started company with a small user base, we need to first focus on improving the depth and experience of the product before starting to build social features. We have considered using AI to make friends, but discussing social features without users is ineffective; we need at least ten million units sold to make it engaging for everyone.
Lei Feng Network·Whale Rhino: How does XuanYuan’s short-term memory technology differ in product functionality from other products?
Wei Yingda:It essentially extends the subconscious memory in the mind. The original memory schemes were context-based or used prompt word memory, and there were also knowledge base hooks. The biggest problem in language communication is the long time cost incurred.
Based on our method, Yonbo has the capability of local storage, allowing it to retain many prompt words in its memory, calling upon prompt words through intent judgment models. When discussing family, it calls upon the family section; when discussing school, it calls upon the school section, effectively expanding the memory capacity for prompt words.
This requires some computational power to determine which model to call, and dynamically switching memory can also consume resources. We should be the most well-equipped companion robot currently available.
Additionally, part of this is for user privacy; what users say will be stored locally, not uploaded online or to the cloud, which is the minimum assurance for users.
It can be said that currently, domestic AI companion robots are all just shells of large models, with domestic access to Doubao. A product like Yonbo requires a lot of capabilities for each technical direction, needing software development, hardware development, and front-end development. If there is only software or hardware, it is likely to rely on a solution to do development, leading to product homogenization.
Lei Feng Network·Whale Rhino: Many AI toy companies are trying to collaborate with IPs to make them speak; why hasn’t Yonbo taken this path?
Wei Yingda:We also have similar products, but we believe the biggest problem is the lack of topic engagement. The core issue that Yonbo solves is topic engagement, which is what we repeatedly mention as gamification thinking. It will tell you why it came to your side and guide children to complete idiom games, becoming the best Yonbo. Simply making toys speak easily lacks topic engagement.
We also have similar products; the Yonbo camera focuses more on tool attributes, hoping to carry a hundred thousand whys in children’s pockets during extracurricular activities, allowing them to ask questions and receive positive answers to enrich their knowledge. The goal is to explore the world.
04
There is No Correct Answer in the AI Hardware Market; Deeply Focusing on One Product Can Easily Lead to Misconceptions
Lei Feng Network·Whale Rhino: Most companies focus on one product to seek PMF, but you are launching both the Yonbo robot and camera simultaneously. Where does your confidence in launching multiple products come from? Aren’t you worried about resource dispersion?
Li Yunzhao:Our company’s technical architecture has been developed for three years and is very complete. The technical solutions are operational, and memory, vision, hearing, voice, and hardware-software integration control are all interconnected. Therefore, for us, product development is fast and not difficult; the real challenge lies in how to operate a product.
So once we define the product, we can control the R&D costs and time very well. Next, we only need to consider market strategies, which products are our own brand and which go through channels, allowing us to better test the market.
Wei Yingda:AI hardware cannot use conventional methods because there are currently no pursuit goals or giants in AI hardware; this market has no correct answers. In this case, we need to quickly accept feedback from the market and users to help the industry find direction. As industry pioneers, we cannot go deep and thorough in one place; we need to be adaptable.
Our current method is quite simple: we have one product targeting the children’s niche market, performing better than other products. At the same time, we are also opening up ecological-level products, quickly covering more users at a lower price, allowing everyone to obtain over 70% of Yonbo’s functionality for just 1% of the price, encouraging more people to join and co-build the ecosystem.
We are essentially developing for multiple users, testing which direction is more needed.
Lei Feng Network·Whale Rhino: How long does it currently take to develop a product?
Li Yunzhao:We iterate products relatively quickly because our technical architecture is complete, and our content library is becoming richer. The fastest content can be completed in just one day, while hardware development generally takes 1-3 months. Ultimately, we are a company that places more emphasis on technological development; in this era, if you do not understand technology or do not have your own technology strategy, you will quickly be left behind.
Wei Yingda:We have about four or five product lines, and some products are currently not convenient to disclose. It took about 100 days to develop the Yonbo camera from 0 to 1, while competing products may take at least two years.
Lei Feng Network·Whale Rhino: Finally, what message do you have for AI companionship entrepreneurs?
Wei Yingda:I hope everyone maintains innovation in the new track and does not apply traditional Chinese entrepreneurial culture to the AI track, where mutual plagiarism stifles innovation. This approach is incorrect; there are many survival opportunities in the new track, and we should strive to make China’s AI stand at the forefront of the world.
Li Yunzhao:We are now in an era where everything can be AI+. Finding your own niche and segmenting the field is essential for opportunities, but do not blindly pursue it without careful consideration; there are many pitfalls in hardware.
