01
A route at a 15-degree angle
To understand CapaLeap's choice, you must first look at what it avoided. The learning machine market is undergoing a brutal round of consolidation. According to data from Runto Technology, in the first quarter of 2026, all-channel sales of learning tablets in China reached 1.252 million units, with the top six brands holding a combined share of 87.5%, 12.3 percentage points higher than the same period last year. Zuoyebang ranked first with a 32.9% share, followed closely by Xueersi at 20.1% and iFlytek at 18.5%. The underlying logic of this track is very clear. Xueersi and Zuoyebang follow a content system—recorded lessons, knowledge frameworks, question banks—equivalent to finding a pre-planned tutor; dictionary pens, repeaters, and word cards follow a tool system, serving a specific function. Both paths converge on the same destination: they are containers for content, serving K12 grade improvement. CapaLeap chose not to squeeze in, but to go around the side. CEO Yang Kun described this strategy with the phrase "a 15-degree angle": the main direction in the market is a matter of consensus, with many players, so competing head-on in the same direction brings heavy pressure and makes differentiation impossible;Keeping a 15-degree angle off the main direction may instead make things suddenly clear and provide additional incremental value.“。If you use an AI Agent for K12, you're just adding one more player to the market, and the incremental value we provide probably can't match the depth of understanding that education companies have of this space.” Not doing K12 doesn't mean avoiding academic content; the difference lies only in how it's taught. Take the Pythagorean theorem as an example: directly explaining a²+b²=c² is a long chain, but AI can offer a visual method suited to a child's age — fill the squares on each of the two legs with water, pour them into the big square on the hypotenuse, and they fill it exactly. Yang Kun said the point is not to hand over the formula directly, but to let children see it and work it out for themselves. In his view, rote memorization and grinding through problem sets are akin to "brute-force solving" — effective for exams, but understanding principles and achieving mastery is what separates top students from ordinary ones. This positioning is what distinguishes CapaLeap from learning machines. By CapaLeap's account, it does not want to be an efficiency tool that completes homework for children, because with an efficiency tool the AI is the subject and the child is passive; nor does it want to be a growth companion, worried that AI could hijack children's emotions and foster dependence; still less does it want to be a teacher, because teachers must give standard answers, whereas it wants children to find answers themselves. Yang Kun positions CapaLeap as a copilot, or a "creation buddy": the child is responsible for thinking and choosing, and the AI assists with execution. Whether this positioning is idealized is another matter, but it addresses a real problem: Generation Alpha are AI natives, and if general-purpose AI tools hand them answers directly from an early age, their capacity for thinking and judgment may be eroded bit by bit. This is precisely the trend CapaLeap wants to resist.02
One brain, multiple endpoints: using an Agent to carry the continuity of growth
CapaLeap's product architecture can be summarized as: one brain, multiple devices, a full ecosystem. The "one brain" is the CapaStudio Agent platform, which provides guided follow-up questioning, task decomposition, knowledge understanding, and creation assistance for terminals of different ages and form factors. Within it, Capa Memory, a growth portfolio, is responsible for connecting a child's interests, expressions, and works formed across different devices and different ages, providing differentiated services based on accumulated memories, cognitive level, and existing input-output records, and continuously accumulating over long-term use. The "multiple devices" form an age-tiered terminal ladder. CapaCap targets ages 3–6, positioned as an AI interest-enlightenment tool that puts the hundred thousand whys into a child's first camera; CapaSnap targets ages 5–9, focusing on AI photo-based learning, emphasizing moving from seeing the world to learning skills; CapaNote, released this time, targets ages 6–16 and is positioned as an autonomous learning and creation terminal. Yang Kun revealed that next year they will also make a wearable-form product, similar to the multi-device relationship between a phone and an Apple Watch. CapaPrint is the companion output terminal, turning digital creations into physical objects that can be held in the hand. Above that is the ecosystem layer. In the user co-creation community, children's works can be shared, used, and iterated upon, but there is no chat-style social networking and no Moments-style social networking. The Co-Leap Program (共跃计划) is open to education, brands and IP, cultural tourism, channels, developers, and content creators. Specific scenarios perhaps illustrate the product logic better. A 4-year-old photographs a pill bug; the Agent first gives a science explanation, then tells a story, then switches to bilingual mode, and finally asks, "Why does it curl into a ball?" One entry is added to the exploration footprint: "crustacean." A child turns a summer pottery class into a small game; the Agent asks back, "How do players win, and what happens if they lose?" After several rounds of dialogue, V1.0 comes out, and after classmates play it, it is revised to V3.0. In CapaLeap's words, the iteration process is itself a learning process. CapaLeap defines this methodology as "Vibe Creating." It originates from "vibe coding," proposed by Karpathy in February 2025, describing creators who are immersed in their own ideas to the point of forgetting the code exists. Applied to children, the division of labor is: the child is immersed in their own train of thought, responsible for thinking and choosing, while the AI is responsible for finding tools and assisting execution, turning curiosity into realizable works.This is also the most fundamental difference between CapaLeap and Q&A-type tools: one hands over an explanation, while the other aims to hand over a work.03
Can the non-consensus path actually work?
CapaLeap has chosen a non-consensus path. Non-consensus means opportunity, but it also means loneliness. According to Yang Kun's estimate, the creativity market is an incremental market: if the top 10% of families nationwide can understand and endorse the product logic of 'using AI to unlock children's infinite possibilities,' that alone would be enough to support a market of tens of billions; if they build something interesting in subject tutoring, capturing a 5% share would also amount to several billion. Whether this estimate is conservative depends on whether the product can turn 'sensory creation' from a concept into a daily experience that both parents and children can perceive. At least for now, the demand is real. Over the past 20 months, CapaLeap's product has entered 220,000 households, children's voluntary daily usage time is about 45 minutes, and 46% of users are still using it after six months. For children's hardware, this retention level is not easy to achieve. It shows at least that the snap-and-learn device is not an electronic toy that gathers dust at home. But the snap-and-learn device's limitations are also obvious. The category's total size is only about 1 billion yuan a year, full of white-label products, fierce competition, and no consensus language has yet formed around user perception. It has validated demand, but has not fully opened up the value space. For CapaLeap to move to higher age groups, it must answer a question: when children are no longer satisfied with snapping and listening, what else can the product offer? CapaNote is an extension in this direction. The snap-and-learn device's interaction leans toward one-way input—snap, recognize, explain—with value that is intuitive but limited. CapaNote, by contrast, lets children use natural language to direct AI to create picture books, podcasts, and mini-games, producing works that can be displayed, shared, and iterated. Once the deliverables are visible enough and impressive enough, the growth logic may change, shifting from reliance on paid user acquisition to word-of-mouth-driven organic growth.This is not a substitution relationship of one rising at the expense of the other, but a necessary step for CapaLeap to move from a single hit product toward a multi-product matrix.The Pai Xue Ji (learning device) handles early-age enlightenment and exploration, while CapaNote takes over creative needs as users grow. The two product lines together form a complete user lifecycle. And CapaStudio's cross-device memory capability is precisely the technical foundation of this stickiness.It turns a one-time hardware purchase into a relationship that keeps accumulating over time.. Users' switching costs rise as their works and memories accumulate, turning data assets into a source of stickiness. Technology can be caught up with, hardware can be imitated, brands can spend on advertising, but a child's developmental trajectory over several years is very hard to replicate. Of course, the risks are equally clear. A non-consensus market needs many players working together to expand the direction, and entering too early may mean becoming a martyr. The pressure of domestic追赶 is already in plain sight: education companies have a base of content and teaching research, and big tech firms have advantages in traffic and models; once the direction is validated, they will not be slow to follow. According to CapaLeap's timeline, 2027 will bring phased results, and 2028 will see fairly substantial improvements. This timetable is neither aggressive nor conservative. It acknowledges that a new category needs cultivation, and also implies that the team has expectations for product iteration.04
Conclusion
Going back to the question at the beginning: when AI can answer in seconds, what do children still need to learn? CapaLeap's answer is: learn to think, learn to judge, learn to create. AI can give an explanation in three seconds, but beyond the explanation, there is observing, doubting, comparing, verifying, overturning, and rebuilding. These actions may look slow, but they are laying the foundation for a person's ability to think. What CapaLeap is betting on is precisely this seemingly "inefficient" path. ItPut AI in the co-pilot's seat, and let the child decide where to go, what to do, and why to do it.. The child voices a vague idea, and the AI helps them find tools, build a framework, experiment, and iterate; when the work is completed, the child gains not just a result, but also the experience of turning ideas into reality. Whether this path can succeed hinges on whether CapaLeap can turn "sensing and creation" into a shareable, accumulable growth experience. If children can continuously produce works they are proud of, parents can see the accumulation of a growth portfolio, and ecosystem partners can derive incremental value from it, then CapaLeap has the opportunity to evolve from a hit-product company into a platform brand. Otherwise, it may just be another brand upgrade after the photo-learning device.As more and more products compete to answer the world on children's behalf, CapaLeap is trying to preserve a space for thinking that cannot be skipped. This path may not be smooth, but the direction itself is already a kind of answer.
(Leiiphone (WeChat account: Leiphone)) (Leiiphone)
Related articles:
A Conversation with X-Origin: The CEO of AI Companion Robots Doesn't Believe in "AI Companionship"
This is an original article by Leiphone. Unauthorized reproduction is prohibited. For details, seeReprint Notice。