CapaLeap Takes a 15-Degree Angle From AI Answer Tools With CapaNote

The backdrop is a growing anxiety about how quickly children outsource thinking to AI. A 2025 survey by the China Youth Research Center of 8,563 primary and middle school students in seven provinces and municipalities found that 61.7% used AI overall and 71.0% used it to help complete homework, while 21.5% said they only wanted to chat with AI and not with real people. A China Youth Daily survey of 1,336 parents found that 58.8% saw signs of psychological dependence on AI in their children and 42.7% felt their children’s independent thinking was declining.

The OECD’s 2026 Digital Education Outlook called the phenomenon 'metacognitive laziness': handing cognitive tasks to general chatbots can create the illusion of mastery while the brain does not process the material. An MIT Media Lab EEG experiment with 54 participants found that a group writing with ChatGPT had significantly lower neural connectivity than a group thinking on its own; after the tool was removed, their alpha and beta networks remained visibly underengaged, according to the report.

CapaLeap is avoiding the main K12 score-improvement track. China’s learning-tablet market sold 1.252 million units in the first quarter of 2026, and the top six brands held 87.5% share, up 12.3 percentage points year on year, according to Luotu Technology data cited in the report. Zuoyebang led with 32.9%, followed by Xueersi at 20.1% and iFlytek at 18.5%. Yang Kun, CapaLeap’s CEO, described the company’s strategy as keeping a '15-degree angle' from the main direction: head-on competition is crowded and hard to differentiate, while a slight offset can create additional value. 'If we use AI agents for K12, we are simply another player in the market, and the incremental value we offer may be completely incomparable to education companies’ depth of understanding,' Yang said.

The company says it does not want to be an efficiency tool that finishes homework, a companion that captures children’s emotions, or a teacher that supplies standard answers. It calls its role a copilot or 'creation partner': the child thinks and chooses, while the AI assists with execution. In one example, instead of reciting the Pythagorean theorem, the AI uses a child’s age-appropriate imagery—filling the squares on the two legs with water and pouring them into the square on the hypotenuse—so the child can see and understand the principle.

The product architecture is 'one brain, multiple devices and an ecosystem.' CapaStudio, the agent platform, provides guided questioning, task breakdown, knowledge understanding and creation assistance. Capa Memory, a growth profile, links a child’s interests, expressions and works across devices and ages. The device ladder includes CapaCap for ages 3 to 6, CapaSnap for ages 5 to 9 and CapaNote for ages 6 to 16; CapaPrint turns digital creations into physical objects. Yang said wearables will follow next year. The company also runs a user co-creation community and an open plan for education, brand, IP, tourism, channel, developer and content partners.

The report gave examples: a 4-year-old photographs a pill bug, and the agent gives a science explanation, tells a story, switches to bilingual content and asks why the bug curls into a ball. Another child turns a summer pottery class into a mini-game; the agent asks how players win and what happens when they lose, and after several rounds the child has a V1.0, which classmates play and help revise to V3.0. CapaLeap calls the method 'Vibe Creating,' borrowed from Karpathy’s 2025 'vibe coding' idea: the child stays immersed in the idea while AI finds tools and helps execute.

CapaLeap says its products entered 220,000 families over the past 20 months; children chose to use them for about 45 minutes a day on average, and 46% of users were still active after six months. The photo-learning device category is about 1 billion yuan a year, with many white-label players and intense competition. Yang estimated that if the top 10% of families understand and accept the idea of using AI to unlock children’s potential, that could support a 10 billion yuan market; a 5% share in subject assistance would be several billion yuan. CapaLeap expects stage results in 2027 and larger improvements in 2028. The risk is that non-consensus markets need many players to build the category; entering too early can make a company a martyr, and education companies with content and big platforms with traffic and models could follow quickly once the direction is validated.