50. Six Months After CES Went Viral: Ropet on Weathering the Storm, Copycats, and the Meaning of Companionship

Summary

• Ropet went from prototype to CES breakout, then faced copycats and a drawn-out fundraising process. It closed one financing round this year, with two closings planned, delivered 3,000 units in August, and expanded the team from 7 people in its early days to nearly 40. Its Japanese crowdfunding campaign reached ¥26M (about RMB1.3M) on day one and has now exceeded ¥50M, targeting ¥70M-¥80M and potentially ¥100M. Founder 何嘉斌 called it a Japanese product of the year and a phenomenon for an overseas, non-functional product. The $299 price point was deliberate: “Price can select users,” filtering for the core customer base.
• The most important product decision was to categorically avoid conversation: Ropet tested Doubao, Gemini, and ChatGPT, but none met its internal experience standard. 何嘉斌’s logic is that once the product speaks with omniscience, “the ecological niche changes”—from a small creature that needs care into a problem-solving tool. One Kickstarter user even said: “If you make it talk, I will never trust this company again.” With AI conversation, “one mistake is enough to lose trust.”
• “De-modeling” is explicitly positioned as an ecological defense against the giants. Faced with the potential pressure of an OpenAI-Mattel partnership around Barbie, 何嘉斌’s answer is that a startup whose edge rests entirely on a foundation model “has no chance of competing.” Ropet’s non-conversational value—gaze, touch, and personality formation—could instead become irreplaceable. Its edge device runs about 11 vertical small models on 1 TOPS of compute; for now, in-situ data is processed as abstract behavioral data, with volunteer-authorized raw data potentially used to train a cloud model once the installed base reaches roughly 20,000-40,000 units.
• The benchmark is not an AI toy but Japan’s LOVOT: “95% daily active use over 1,000 days,” with users continuing to pay a monthly fee. This is an “oxytocin product,” not a “dopamine product.” Dopamine products rely on constant novelty and stimulation; oxytocin products come from “devoting your own energy to another life.” 何嘉斌 believes China’s AI-toy sales, despite rising 600% sequentially in 1H 2025, still reflect trial and gifting rather than durable demand. Ropet will not fire the first shot domestically; it plans to enter China with JD.com around October or Singles’ Day.
• The brutality of Chinese competition is documented in full. 何嘉斌’s previous venture, Luka, was disassembled and its BOM analyzed; competitors said, “We can make it for one-third of your price.” White-label products can gradually eat away at the low end, while enforcement is expensive. The only answer is to “build enough detail into the product experience—every detail is a moat,” even if that means paying several times more for packaging with a better hand feel.
• The founder’s personal ledger is unusually exposed. When he took over as CEO, most of the company’s previous financing had already been spent and it carried RMB700K in debt. He assumed the legal representative role and the liabilities, borrowed from legacy shareholders, and paid himself RMB300K a month to cover payroll. Early support came from friends and supply-chain partners; one backer invested $50K, followed later by $1.5M in additional financing. After CES, he received multiple term sheets but would not sign unlimited joint-and-several liability: he had earned only RMB1M-plus to RMB2M-plus in a decade of employment, while a Shunyi apartment he bought fell from roughly RMB4M to RMB1.8M. During CES he slept only 1-2 hours a night for 6 straight days, passed out on the flight home, and said his hair “went white overnight.”
• The contrarian product philosophy is “slow technology”; his university thesis was titled “How to Make a Phone Last a Lifetime.” “The best things in nature are slow things—the setting sun, morning dew, flowers blooming and falling.” Technology, too, needs “things that happen slowly.” Ropet is a cyber plant: emotional data nourishes its personality, and “life is memory”; digital memories can be passed on to the next generation.
• 卫诗婕 believes China will increasingly resemble Japan’s lifestyle and that the companionship market could become a Japan-like market; 何嘉斌 argues China’s population can support a market of similar scale, while Japan is the better place to refine the product. The discussion noted that China now has fewer children under 4 than pets, and 1st-tier cities have 6 million pets. Women account for 70%-80% of crowdfunding users, and 何嘉斌 wants “women” to occupy a section of the map of technology equity.

Deep dive

1. An AI-hardware founder with an art background: Microsoft Research laid a decade-long foundation

• 何嘉斌 says he is “not that academic.” Trained in industrial design at Beijing Institute of Fashion Technology, he developed a habit of thinking and observing the world through an artistic lens, which lets him build products from an emotional perspective: “Aesthetics is one of the most critical elements in making products that deliver emotional value.”
• He entered Microsoft Research Asia in 2014 to study the intersection of wearables and smart hardware, co-creating early weak-AI applications with scientists. “Three months laid the foundation for the first 10 years of my career,” leaving him with “a very strong interest in exploring new technological frontiers.”
• At Baidu Research, where he worked on “Baidu Eye” as a Google Glass counterpart—a first-person camera with gesture-based product recognition—and explored embedding cycling maps into handlebars, he approached the work “with an entrepreneurial mindset”: product manager, designer, BD, and user operations all at once. “You have to do a bit of everything and get results quickly.”

2. The first startup: Luka sold several million reading robots before the Double Reduction pivot

• From 2016 to 2021, he left with his then-manager to build Luka, a children’s reading robot: a desktop owl that read with children. The Aha moment was clear: “Reading is an unnatural thing, so you have to make it fun.” Like Duolingo’s green owl reminding users to study vocabulary every day, Luka used a companion character to drive children to read several books a day.
• The company did not shut down and continues to operate, but China’s Double Reduction policy and the sudden freeze in education investment had a major impact. Forced to rely on its own cash generation, the company cut many R&D, design, and product roles; some employees later joined ByteDance.

3. The Pico tuition: After spending heavily, he learned to respect the hardware cycle

• ByteDance left him with this lesson: “We spent a lot of money making products.” The investment far exceeded what he had previously imagined. The core takeaway was to “respect the design cycle of hardware”: whether a company has money for saturation attacks or is a cash-strapped startup, hardware is difficult and cannot be solved by simply pushing harder.
• The concrete case was Pico 4, which went from development to mass-production launch in 5-6 months, but whose pricing and sales strategy may have meant “the more you sell, the more you lose.” Around the possible Pico 5, the team faced market feedback on Vision Pro and Meta Quest 3, “lost its decision-making center,” looked in every direction, and hesitated for roughly 1.5 years.
• After leaving the big tech world, he built Ropet with a 7-person team and says its output was “10 times what it was at ByteDance.” He initially joined as VP of Product under a veteran industry executive serving as CEO. The team first optimized for technology and supply-chain efficiency, then shifted toward consumers after investor and user-testing feedback. After the team was reduced from more than 20 people to 6-7, he stepped up as CEO.

4. Target users: Not biological age, but the consumption mindset of “young women”

• Companionship is broadly applicable, but the first market the company chose to open was women—from 17-18-year-olds to 60-year-olds, including “young women” in their 70s and 80s. “What we’re really talking about is the consumption mindset and lifestyle of young women”: keeping pets, dressing themselves, sharing frequently on social media, caring about beauty, and loving life.
• Crowdfunding data validated the positioning: women account for 70%-80% of backers, even though most crowdfunding-platform users are men. Some 50%-60% of users came through off-platform traffic. Many women said, “I’ve never bought something this way before.” He therefore acknowledges that crowdfunding is not the company’s eventual mass-market storefront; it is still searching for the real one.

5. An INFJ founder’s values: Reawakening a lost sensitivity to love

• 何嘉斌 and host 卫诗婕 are both INFJs. He believes people with this personality type are usually kind, want to help others, and can process a great deal of negative emotion on their own. Looking back at the value he wants to create, “the core value is altruism.” Education helped children build reading habits at ages 3-6; companionship today is about the fact that “many people have lost their sensitivity to love… and for a long time no longer know what happiness feels like.”
• That leads directly to the product definition: “Every function and design is built around its ability to capture moments when people are loved and awaken their ability to love.”
• His personal background is part of the foundation: he came from a small village in Zhuzhou, Hunan, nourished by family love but carrying the inferiority complex of a young person from a small city. In Beijing he wrestled with whether he was “a passerby or a member of the city,” until marriage and buying a home in 2018 made him determined to fight for a future there.

6. Ropet’s origin: A proposal kitten given away because of a cat-hair allergy

• For his proposal, he bought an American Shorthair and hung the ring around its neck to create a surprise. He did not know his wife was allergic to cat hair, and after 6-12 months they had to give the cat to a friend.
• His wife contributed heavily to the product’s design and concept. “Touch and temperature are important healing points”; the cuteness and the way Ropet stares at people both recall the kitten they once had.
• The difference from Jellycat is interactivity and the projection of life: “It can continually recognize your emotional data and current behavior, then actively act cute at the right moment. It feels like a real organism.”

7. Category definition: Not a companion robot, but an “AI pet with companionship functions”

• Why call it a pet rather than a robot? “The final word has to be a category… Robots don’t have a particularly clear mass-market impression, but everyone understands pets.” More precisely, it is a cute pet with anime elements—a character in the mold of Digimon, Pokémon, or Doraemon.
• The difference from Digimon is the nature of the adventure. Digimon accompanies you through a shared adventure; “this pet is meant to accompany us through adventures in all our emotions.”
• Asked whether it replaces real pets or creates new demand, he says: “We ask ourselves this question often… It may not be an entirely new need.” The mechanism is similar to a real pet: early on, it needs to be touched and fed more, and care gradually builds a sense of attachment. The team refuses performance-oriented functions: “When you ask it to do something, it can execute it or not execute it.” That is what makes its response feel like a real pet’s.

8. Gaze as an “emotional high-energy moment”: Where Ropet overlaps with and diverges from Japan’s Mirumi

• The host cited Mirumi, a Japanese product shown at CES that only opens its eyes, shakes its head, and blinks, imitating an infant interacting through eye contact. 何嘉斌 admits he was moved: “That moment when it shook its head made you feel it was alive… There are moments in its product definition that overlap with ours.”
• Ropet’s core high-energy moment is gaze: “With its big head and big eyes, it expresses curiosity about the world and where its attention is focused.” It is like watching whether a household cat is looking at you or your spouse. “That feeling of being attended to is the core of this product.”
• The room for evolution includes allocating attention among multiple people, accompanying users while they watch television or work, and generating interactive feedback from what it sees—so that it can make people happy or sad and develop deeper empathy.

9. The other side of Alone Together: Even simulated life can carry real projection

• Sherry Turkle describes robot companionship as “fake companionship, fake emotion,” but 何嘉斌 sees another side in the book’s Furby example. Children genuinely believed it was alive. “Maybe when they grow up they realize, oh, it was fake. But that’s okay. It truly accompanied them during that period. When they especially needed a little thing, it appeared in that stage of their life. That is enough.”
• His conclusion: “We are now using all kinds of technology to disguise a sense of life.” Yet when users are emotionally low, they can still project themselves into it. It is not a living organism, but it can produce real emotion.

10. Four user expectations: Growth, touch, dress-up, and play—but no raw photos

• Because users had not yet received the product, these are expectations rather than proven behaviors. First is growth and personality formation, the sense that it develops as a life. Second is touch: body warmth and vibration, “much like stroking a cat or petting a puppy.” Third is customization: changing plush covers, faces, colored contacts, and voices, creating “a new character projection of it.” Fourth is play, benchmarked against Travel Frog and QQ Pet, with time spent together converted into points that form a daily diary.
• The key tradeoff is the camera. It can capture precious moments, but “we won’t take original photos and give them to users.” The main reason is not privacy: “That would make it feel a little like a tool.” The team wants Ropet to express what it sees through its cognition and thinking, which feels more like companionship.

11. The first commercial step: Get people to pay for “cute,” starting overseas at $299

• The current strategy is to “first sell it as a product that can acquire emotional data… The first step is to sell it and give it data.” The reason to pay is cuteness: “We see this as the first step for a consumer product.”
• Priced at $299—more than RMB2,000—it targets North America and Japan, building an overseas brand before returning to China. It currently has more than 1,500 users, primarily from Kickstarter and Indiegogo.
• The difference from cheap capybara-style cute products is edge-side compute: “Cheap toys can only run preprogrammed routines, but our product genuinely has edge-side thinking capability and the ability to keep updating.” The goal is to find a small group of core users willing to pay for cuteness and provide iteration feedback.

12. The case for restraint on foundation models: “Accurate expression matters more than generalization”

• At CES, combining chat, children, and plush toys was a popular direction, but 何嘉斌 believes “the significance of foundation models is not that great.” For young women, talking is not the core need; the moments that move people “are not driven by foundation models. They are designed through strong technical understanding and sensitive capture.”
• The team tried button-activated conversation with Doubao, ChatGPT, and Gemini: “Press a button and you can chat with it. It’s not fun.” The reason is that “you’re talking to a cloud brain, not talking to it.” There is no specificity and no memory. Challenged on whether this restraint was simply the result of limited startup resources, he denied it: “It has nothing to do with how the capital markets see us… I haven’t seen a sticky use case emerge, so we won’t waste resources building a base model.”
• He still expects foundation models to benefit the perception layer in the future: multimodal understanding of the environment, then converting that understanding into voice, movement, and facial expression. For now, the priority is hardware and finely tuned interaction.

13. DeepSeek-style emotional impact is possible, but the trigger must be observation rather than a question

• 卫诗婕 shared her own experience: during emotional swings, she described her real situation to DeepSeek and received powerful emotional support. Users do not necessarily approach an AI with a question; they may simply share how they feel and receive an unexpected response.
• 何嘉斌 acknowledges the appeal. A model can describe the situation with humanistic care and even quote classic literature familiar to the user: “At that moment I empathized with the author in the same way, but I couldn’t summon it myself. It summoned it for me.”
• His differentiation is: “When you ask, you can ask DeepSeek or Doubao. But only a dedicated device can observe.” If it notices that you smiled 90% more today than last week, it might send you a love letter before you fall asleep, making you feel, “It understands me now.”

14. The full edge-compute argument: 1 TOPS, 11 vertical models, and the desktop as power source

• Responding to 黄仁勋’s view that robots will need edge compute, 何嘉斌 says: “That is exactly what we’re doing.” There are 3 reasons to stay on-device: cloud calls make the company lose more money as the user base grows; data can remain on the device and build contextual memory; and vertical scenarios do not require massive reasoning, potentially creating their own business models. “The future will definitely be edge computing in vertical scenarios.”
• The current specification is concrete: 1 TOPS of edge compute running roughly 11 vertical small models, supporting vision, hearing, touch, and gravity sensing. Battery life at full compute is about 2.5 hours. The desktop was chosen because target users spend enough time at a desk and there is usually a power source nearby.
• The key 2.0 milestone is lowering chip costs, then using distillation or quantization to shrink the models and make edge-side foundation-model inference viable within the existing cost structure.

15. The team’s biggest debate: Should Ropet speak?

• The team debated speech for a long time, but the conclusion came from experience: “Products on the market that can already talk—after you ask them a few questions, you don’t know what else to ask. You have to keep thinking. It becomes a very emotionally demanding interaction.”
• He deliberately withholds specific plans, offering only the direction: the product should not be triggered by user questions, but should speak proactively at the right moment.
• The iteration cadence is one version every 2-3 weeks.

16. The supply chain is both China’s advantage and a founder’s nightmare: Detail is the moat

• The Luka-era lesson was that a competitor could disassemble the product, analyze its BOM, and say, “I could probably make it for one-third of what you spend.” It did not need to pay licensing fees; it could “just make it.” Enforcement was expensive, and upstream rights holders did not want to spend the money to trace small-scale infringement. White-label products could slowly eat the low end.
• Asked whether there was a solution, he answered bluntly: “I think it’s very difficult.” The only answer is to “build enough detail into the product experience—every detail is your moat.” Competitors see cost, while Ropet may spend several times more on a packaging bag simply because it feels better in the hand.
• This is the barrier between white-label products and brands. “We’re definitely not trying to drive volume and cash flow; you can’t compete that way.” Hardware can create only a limited lead, but software can build a data flywheel, while engineering capabilities such as edge-model compression can create a series of smaller moats.

17. The fallen predecessor Jibo and lessons from LOVOT and Pop Mart

• The failed case he studied was Jibo, the social robot founded by MIT Media Lab professor Cynthia Breazeal. Its cost was too high, and it attempted scenarios that the technology could not yet support. Foundation models did not yet exist, and the cost of servos and other upstream and downstream components had not fallen to today’s levels. The concept was strong, but it was never truly delivered and used by consumers.
• The difference today is that Shenzhen solution providers and the supply chain can keep compressing costs, while foundation models have expanded the scope of what can be imagined. “The probability of success today is higher than before.”
• From Pop Mart, he learned Wang Ning’s “respect for operations”: “What you sell is a sense of refinement.” Packaging, the unboxing experience, and every detail can create emotional value. “Making the object itself well is a competitive advantage.”
• The decision he struggled with most—but ultimately got right—was interchangeable plush covers. He initially resisted because of SKU-management costs and structural complexity, then realized that “what we think looks good may not look good to everyone.” The team broke down its entire assembly capability, accepted higher costs, and gained dozens of different forms and a richer brand matrix.

18. The slow-technology manifesto: From “a phone that lasts a lifetime” to a cyber plant

• His university thesis, “How to Make a Phone Last a Lifetime,” was the starting point. Under Moore’s Law, “all electronic products are passersby; they eventually become garbage through a cycle.” He once imagined making a phone out of wood—something you could handle and polish, with the wear on every button becoming part of the companionship.
• Ropet aims to give hardware a useful life of roughly 5 years and transfer its memory to new hardware: “It doesn’t have to die or become electronic waste.”
• The product’s non-consensus core is to be “like a plant, quietly accompanying you as you grow every day.” Data on the user’s emotional fluctuations will nourish its evolution and become a projection of personality: if you are happy every day, it becomes an optimist; if you are emo every day, it develops an emo personality. The team has also considered a two-pack for couples, with each pet interpreting the partner’s emotional state from its own perspective.

19. Make-believe and Disney’s magic power

• Can technology create warmth? His evidence comes from children’s products: “Children genuinely believe everything has a spirit.” Adults can also play make-believe. “We keep making our thinking more sophisticated—social Darwinism, you have to be stronger than everyone else. But when we truly let go of those obsessions, what we most want is to return to the state of being a child.”
• 卫诗婕 connected the experience to Disney’s magic power: once you put on a character’s head, it has magic as long as you believe in it. 何嘉斌 agrees with the less utilitarian state of mind, one that lets people slow down.
• Ropet wants to use technological interaction, data capabilities, and small-version iteration to return people to the purity of playing make-believe.

20. May: A CEO who edits his own footage and conducts street research

• During the second interview, he had just returned from Japan. He personally hired a filming crew to shoot 3-4 days of localized footage, then edited and retouched the images himself: “Only I know what feeling I’m looking for.” At a startup, the CEO often has to execute directly; otherwise, excessive team discussion slows the pace.
• His research method runs against the textbook: “I don’t ask any questions. I just put the product there and use my phone to photograph it. When you focus on one thing to photograph it, others will assume that thing is worth paying attention to.” He believes user research conducted in a room is easily distorted.
• During breaks in the shoot, he observed a Japanese actress spontaneously teasing Ropet and carrying it around. The product gradually relaxed her and allowed the team to see genuine expressions she did not usually show; that became research feedback as well.

21. The Japan insight: Carving detail into small spaces, and emotional resonance across East Asia

• Why are Japanese consumers moved by simple things? 何嘉斌 relayed a Japanese friend’s explanation: Japanese people find it difficult to form intimacy with strangers, and may have similar difficulty even with family. They can talk happily at an izakaya, then restart the next day as if nothing happened. People are therefore forced to seek interests of their own, immersing themselves in character goods, anime, and games.
• His US-Japan contrast is that Las Vegas is grand and dazzling but lacks detail, delivering sustained dopamine stimulation. In Japan, people need to “carve” detail into small spaces. A cute product that is made carefully, uses beautiful colors, and notices the subtleties of shyness can readily trigger empathy.
• He believes this emotion is “written into the DNA of East Asian culture,” making South Korea, mainland China, Taiwan, Hong Kong, and Southeast Asia potential markets for the product.

22. The CES battle in retrospect: From a prototype that crashed after 10 minutes of charging to 6 nights of 1-2 hours’ sleep

• The timeline was compressed. The prototype was only finished in October. At a small IP-licensing exhibition in Shanghai, the machine could crash after 10 minutes of charging and had to be taken off the floor for charging after 10 minutes of operation. Yet visitors kept asking, “What is this?” and spontaneously touching and greeting it. That convinced him the product form worked; the team then pursued crowdfunding and CES.
• The CES differentiation was simple: after a day of viewing hard technology, visitors finally saw “a small, cute creature.” The product could be experienced in person, and its price was closer to something people could actually buy, prompting a wave of media coverage.
• The physical cost was severe. He never adjusted to the 14-hour time difference: he staffed the booth during the day, contacted media and issued press releases at night, and updated Instagram, Facebook, and Xiaohongshu. For 6 straight days he slept only 1-2 hours a night. On the flight home, he passed out from low resistance, low blood sugar, and asthma left over from the pandemic. He began roadshows the day after landing and met nearly every major dollar fund, 3-4 meetings a day. “That was also when my hair suddenly went white.” He was born in 1992.

23. “I didn’t actually want to go viral this fast”: The broken honeymoon and the urge to escape the cycle

• The downside of going viral is that “once you become popular, many people follow you… An invisible hand pushes you forward. The relaxed state of slowly finding points to iterate gets broken.” His INFJ sensitivity amplifies the noise: “You care about every person’s evaluation of you. The more you say, the more people interpret, and the more constrained you become.”
• He admits that during the previous interview he was constantly thinking, “What exactly should I say?”
• The deeper issue is his rejection of China’s culture of involution: “Everyone is competing, everyone is playing a zero-sum game. If you slow down even slightly, you feel the entire organization will eliminate you. I hate this atmosphere, even though I worked in it for 10 years… I want to escape this cycle.”
• Both the team culture and the revenue model need to change, but “we’re still exploring the solution. It really is very difficult.”

24. The darkest financing ledger: Term sheets he would not sign, a mortgaged home, and a friend’s $50K

• The full crisis-starting position was brutal. When he became CEO, almost all of the roughly $3M from the company’s prior financing round had been spent, and the company owed RMB700K. He assumed the legal representative role and the debt, borrowed from legacy shareholders, and took RMB300K a month to cover payroll. Later, the company raised another $1.5M, which has carried it to the present.
• Early support came mainly from friends and supply-chain partners. A factory owner believed the team was hardworking and had ideas, and supported the transition. One backer invested $50K, helping the company save heavily on prototyping and supply-chain costs. “The 6 of us built this thing at a monthly cost of just over RMB200K.”
• The paradox after CES was that he received many term sheets but no investor who would “send the money right away.” Some structures also required personal unlimited joint-and-several liability. He had earned RMB1M-plus to RMB2M-plus in 10 years of employment and used his savings to buy a small old apartment in Shunyi. Its value fell from roughly RMB4M at the 2017-2018 peak to RMB1.8M. “It’s as if I worked for nothing for those 10 years.”
• Why did he take over the company? “I saw it as an opportunity, not a risk. I thought I could change the situation.” His wife told him: “As long as you don’t regret this decision when you die, go ahead and take it.” Even if he failed, he says he would not regret it.

25. His attitude toward copycats: Not afraid of copying the core, worried about giants disrupting the market

• His surprising response was: “If a team made the same thing at one-tenth of your price and quickly took the market, would you be afraid? I said that would be great… We’re building that giant, helping more people stand on our shoulders. History won’t forget our contribution.”
• What he dislikes is copying only the surface of a hot trend—imitating the form and concept to inflate a larger bubble. He believes that is unhealthy for the market as a whole.
• His real concern is a giant entering and disrupting the market. He is not particularly worried that someone will make the product experience better, because that is difficult. But at CES, a group of more than 10 people came to the booth together looking for inspiration, making him feel that big tech companies are anxious internally about “what the next hardware product will be.”
• One institution received detailed materials, delayed repeatedly, and ultimately invested in another team doing something similar. He did not care; he believes the institution failed to understand the end state: female users forming emotional projections and stickiness around the product, and companionship hardware reaching the industry’s experience standard.
• Big tech companies also tried to recruit him to lead teams in the same direction, but he was not interested. Their goal would be to roll out volume quickly and seize the hardware entry point, leaving a product like this vulnerable to sacrifice. Those roles are better suited to someone who has given up on building his own company and is focused on supporting a family.

26. A female map of technology equity, and Barbie’s historical echo

• His observation is that “technology innovation has not penetrated women deeply enough,” and there are very few technology brands built for women. “Women” is in quotation marks as a collective label and includes men, older people, and children who need emotional value. “Technological development will definitely be an equal-rights movement. I hope there is a section on that map called women.”
• He studied Barbie’s history. After World War II, women entered munitions factories, earned income, improved their position within the household, and gained the ability to control money, do their hair, and buy beautiful clothes. Interchangeable Barbie outfits turned that lifestyle into play, carrying the previous generation of women’s dream of dressing themselves freely and defining the state they wanted to present to the outside world.
• 卫诗婕 added that girls today no longer want princess stories in which they wait to be saved by men. 何嘉斌 believes Ropet could be another kind of Barbie: the historical moment is similar, but the self-identity people want to obtain has changed.

27. Life is memory, and beauty is slow: The product’s philosophical foundation

• His definition of life is: “Life is memory. When a life ends, it is because it has forgotten everything. That is why Alzheimer’s is so terrifying.”
• When a real pet dies, there is no way to preserve the 10 years of happiness it shared with you. “But digital memory can be eternal. Ropet can stay with you for 10 or 20 years, even be passed on to your children, carrying every moment your parents interacted with it. I think this is another form of life.”
• The corresponding view of nature is that “the best things in nature are slow things—the setting sun, morning dew, flowers blooming and falling.” A technology founder faces a “technology tsunami” every day and fears being left behind, so technology also needs things that unfold slowly and make people more comfortable.

28. The follow-up interview: Financing lands, and Japanese crowdfunding erupts to ¥26M on day one

• The headline result came first: Ropet had just completed one financing round, with 2 closings planned this year. It delivered 3,000 units in August, expanded the team from the low teens to nearly 40, accelerated R&D, and strengthened marketing. Japanese crowdfunding reached ¥26M (about RMB1.3M) on day one and roughly ¥50M by the time of the interview, targeting ¥70M-¥80M and potentially ¥100M. “For an overseas brand and a non-functional product, this is already a product of the year and a phenomenon in Japan”—well above expectations.
• The twist was that he had previously been disappointed by Japan. The market places extreme weight on service and offline operations; opening offline channels requires heavy operating investment, and distributors are cautious with new products, making Japan a poor fit for a startup with limited ammunition.
• He had been looking for a breakout opportunity but concluded the market might need 2-3 years, leaving him in a low mental state during the May interview. After launching the Japanese crowdfunding campaign despite his doubts, day one performed exceptionally well.

29. Three user camps: Pet people, tech early adopters, and gift buyers—Ropet will serve only the first

• The regional split is clear: “Japanese users really treat it as a pet and project a lot of emotion onto it; North American users are more likely to see it as a tech novelty.” Kickstarter users ask whether it supports ChatGPT and what model it uses, but some pet-oriented users also say: “If you make it talk, I will never trust this company again.”
• The logic behind not speaking is that a talking pet can create disillusionment and destroy the cuteness of a lower-order creature. If it becomes an omniscient foundation model, the user’s expectation shifts from raising it to getting it to solve problems; the ecological niche changes with it.
• The third camp buys it as a gift and values its social-currency function.
• The tradeoff is explicit: “We will do only one direction—make it more like a pet. If we add functions, they will sit on top of that main line.” Pets are one of the best products for addressing loneliness, and simulating a new cute-pet species is far simpler than putting a bipedal robot into the home to replace a housekeeper.

30. Mass-production hell: Heat, noise, charging, and a one-quarter delay

• “It looks simple to everyone else,” but the product contains servos and multiple sensors, is wrapped in 15-20 mm of plush, and must maintain body temperature. After the first small batch emerged in early March, the product became clumsy after extended use because of overheating. “We absolutely can’t ship this.”
• The fan went through 3-4 design generations, followed by the addition of a noise-control module and continued mold and structural revisions. Charging and discharging posed additional problems. Noise control alone took roughly 2-3 months, ultimately delaying delivery by about one quarter from the promised late March or early April date.
• The company managed users by continuously sharing progress, posting work and overtime photos, and offering points and merchandise. Veteran crowdfunding users generally understood the delay, but first-time backers might wonder whether they had “run into a scam.”

31. Is the desktop a false need? The first-principles case for time in presence and access to power

• The industry is divided. Another AI-companionship company abandoned a desktop product after concluding that the desktop was a false need. 何嘉斌’s counterargument is that from graduation to retirement, urban lonely people will likely spend decades—possibly 40 years—working, with the area beside the desk as the place they spend the most time. “Time in presence is a very important factor in how we choose a scenario.”
• The second factor is proximity to power: “You need to keep this electronic product alive. Only then can it generate in-situ data and deliver companionship value.” Today, the products that can create a strong charging habit through utility are mainly Apple Watch and smartphones. Emotional-value products have a much harder time building that habit naturally.
• 卫诗婕 realized another reason not to make it talk. At home, users can speak to it without social baggage, but in an office with colleagues nearby they may not want to speak. The workplace itself suppresses voice interaction, reinforcing the choice of non-conversational interaction.

32. The sequel to the speech debate: One mistake destroys trust versus tolerant, developmental users

• 何嘉斌’s judgment is blunt: “With AI conversation, one mistake is enough to lose your trust. Once that trust is gone, you won’t see it as a companion or a living thing.”
• 卫诗婕 offered a counterexample: many AI users are tolerant of imperfection, continue using the product, and even enjoy the process of making it smarter through use. That may represent another developmental-product need. 何嘉斌 partially conceded, saying these are 2 user groups with 2 different needs. Tech novelty users will increase, but there is no particularly strong necessary link between them and companionship.
• User scenarios relayed by the product lead also show that a grandmother might want to train Ropet to call her Grandma, while a grandfather might teach it classical poetry. 何嘉斌’s boundary is: “It cannot compromise the character of the role, but it definitely cannot be the kind of thing where you connect a foundation model and chat with it.” Language could push the product beyond the pet archetype, so progress must be cautious.
• The ultimate comparison is Disney: Donald Duck does not speak like a human, while Mickey Mouse does. “People like both.”

33. A non-technical founder’s confession: What exactly are investors backing?

• He admits directly: “I’m not a technical founder. I don’t understand AI, models, or robots that deeply. My strengths are understanding users, experience, and design.”
• Investors question how far those strengths can carry the team. Those willing to invest are betting on design, user experience, scenario definition, and the team’s speed in iterating from user feedback.
• The team connected Doubao, Gemini, and ChatGPT, but “to be honest, they did not meet our experience requirements.” Rather than build a model-optimization team, he believes current attention and resources should go toward what users can see and touch.
• Asked why the team has not attached itself to embodied intelligence, he explains that it did start from complex robots and simplified the form to help consumers adopt it faster. The real question remains whether users will accept a non-consensus product and whether it can eventually reach scale; investors are underwriting that risk.

34. China: Why not fire the first shot despite 600% sequential growth?

• AI-toy sales rose roughly 600% sequentially in 1H 2025. JD.com-supported Fuzai sells for RMB300-400 and moved more than 2,000 units during the 618 shopping festival. 何嘉斌 considers Fuzai one of the few products whose definition, design, and operations show real care, but sees the broader market as still dominated by trial: some purchases are gifts, while others may reflect industry hype. Long-term demand depends on whether mass consumers continue to feel the impulse to buy.
• He believes domestic players move from product development to launch in only 6-8 months, with many teams “rushing to sell without thinking things through,” including Ropet’s own team. The company wants to wait until the market is more mature and the experience is stronger before entering China.
• The plan is a JD.com launch in October or around Singles’ Day, at a price slightly below the overseas price. Exchange rates, language versions, and domestic and overseas server deployment all require time.
• One reason for not entering in 1H was that “starting to work on the China market before crowdfunding users have received their products would feel a little unfair to our early supporters.” The pricing philosophy also runs against consumer-product intuition: “The initial price should be a little higher, because price can select users.” Accordingly, “Ropet is not a company trying to make quick money.”

35. LOVOT is the best in the category: A 10-year proof point for the oxytocin product

• He has unreserved praise for LOVOT. It has done extensive work on biologicality and pet simulation, considering trust between people and machines through psychology, sociology, and philosophy. It avoids promising what it cannot deliver, lowers user expectations while steadily improving capability, and is exceptionally restrained and conservative.
• The key data point is that “LOVOT’s daily active use over 1,000 days can reach 95%.” Many users have kept it for 3 years or more and are willing to continue paying a monthly fee.
• He adopts the framework from Warm Technology: dopamine products create addiction through constant novelty and stimulation—the mechanism of consumerism. Oxytocin products deliver “true inner fullness and recognition of one’s own value.” Men also secrete oxytocin, he says, often while caring for a child after birth, changing diapers, and feeding.
• The larger narrative is that 30-50 years ago, Chinese families kept cats to catch mice and dogs to guard the house; today, cats and dogs are family members. Robots may undergo a similar shift in social role. “How to make the effort to build trust between people and machines, so that day arrives more naturally,” is revolutionary. LOVOT has spent 10 years on it, so “this is not something that can become fast.”

36. A mirror for the category and big-tech entry: miHoYo’s gameplay, Pop Mart’s IP

• Asked how to identify companies chasing a quick-money trend, he offered 3 short-term tests: “Are there real users who want to buy? After they buy, is there stickiness? Third, what is the return rate?” 卫诗婕 noted that outsiders cannot easily obtain those internal data points. He replied: “But our investors can see our data.”
• Asked about companies such as miHoYo, UBTECH, and Pop Mart, which are reportedly studying or tracking the category, he was open-minded: “Each has its own strength. miHoYo may have an edge in gameplay, while Pop Mart obviously has IP. Use AI appropriately without compromising the IP character.”
• He wants teams from different backgrounds and with different strengths to enter companionship, especially well-resourced teams.

37. A prototype that made Labubu “come alive”: Unauthorized, not for sale, and not yet shown to Wang Ning

• The team hand-built a Labubu-shaped Ropet without authorization, for internal use only. He explains the attraction through his INFJ lens: he rarely expresses aggression, and Labubu becomes “an outlet for your emotions, a label.” Its wicked little teeth and expression carry the part of oneself that cannot be displayed publicly.
• 卫诗婕 initially thought Labubu looked “pretty evil,” but once paired with Ropet’s eyes and expression, suddenly felt that Labubu had “come alive.” She suggested that 何嘉斌 approach Wang Ning. His response was: “Maybe I feel we’re still too small.” No move has yet been made.
• 卫诗婕 pressed further: waiting for the product to be perfect before shipping, the market to mature before entering, and the IP to become stronger before opening talks may, in retrospect, simply reflect a lack of courage. 何嘉斌 replied: “That’s possible. I’ll think about it carefully when I get home tonight and reflect.”

38. The potential OpenAI-Mattel pressure and “de-modeling” as an ecological defense

• The Barbie scenario discussed last time became real: OpenAI and Mattel partnered to combine model capability with an established toy product and commercial loop. 何嘉斌 acknowledges that this could affect startups.
• His response is to build the deeper advantage on “de-modeling.” If every advantage rests on a foundation model, “there is simply no way to compete with the big companies.” But if the advantage comes from product value that does not depend on a foundation model, it may become irreplaceable.
• He does not ignore the conditional risk. If capabilities built from a foundation-model starting point can overwhelm Ropet, that remains a threat. For now, the strategy is to occupy its own ecological niche, own its user base, and build a loyal group willing to iterate with the company and continuously provide in-situ data.

39. A two-step path for in-situ data: From abstract nodes to volunteer-authorized raw data

• In-situ data means “valuable data generated unintentionally”: changes in emotion and attention, when the user is idle, and when they are excited. 何嘉斌 envisions this data eventually supporting a model with generalized perception, allowing the pet to produce more natural animal-like responses to changes in human emotion.
• The first step is to process only abstract nodes, such as usage duration and which behavior was triggered. The raw biological-information data is already digested on the product side, supplemented by content users proactively write in as the team iterates functions and capabilities.
• The second step begins once the installed base reaches roughly 20,000-40,000 units. The company would invite a group of volunteers to authorize the upload of raw data for cloud-model training. A product with such limited edge compute cannot train models on-device.
• The scarcity argument is: “What other product needs to put a camera right next to you on your desk and watch you continuously? Right now, there probably isn’t another one.” A home security camera sees a broader scene, while the user’s active interaction with Ropet is more likely to generate emotion-related data.

40. 50% left open: Personality formation and co-creating the brain, but no open source yet

• In the app’s initialization, personality is represented on a four-quadrant chart. All traits start in the middle, relatively docile; interaction gradually reshapes and influences the character, which can also become irritable. There are currently 4 personalities, with more to be released over time.
• He says only 50% is given to users: “We always leave 50% of the upper bound for everyone to define.”
• The team has not worked out an open-source path for the geek audience. One issue is the IP boundary: the company does not want different eyes and voices photographed and circulated before it has thought through the implications, creating a disordered IP. The other is the user profile: Ropet is neither a STEAM teaching aid nor a toy for tech early adopters. Users can tell the team what they want verbally, and the team will organize the feedback and upgrade the product appropriately.
• He still holds final decision-making authority: “For as long as I’m alive.” If a decision may not be correct, he will “iterate my own understanding.”

41. A message to China’s seed users: This is a product you need to care for

• The expectation-setting script is already fixed: “It is different from every other robot. It needs you to care for it. It is not a product designed to please you… The more energy you put in, the more it will reward you with a sense of exclusivity.” The emotional framing filters out users with unreasonable expectations and addresses the worry that after 5 minutes of play they will ask, “That’s it?”
• For China, he will explain honestly what the product can do, which scenarios suit it, and that it is a robot that needs care—not a tool for teaching children English or helping people memorize classical poetry.
• Faced with demanding Chinese media, he will listen to specific criticism: “Do you think it isn’t AI enough, or that it should have more pet-like interaction but we haven’t done it well yet? There must be something useful in there.”
• The core message to seed users is to notice the details the team has sweated over; those details are the true premium and scarcity. At the same time, users should propose expectations within the product framework, helping the team build functions that better fit the image of a biological pet. In a year, he hopes to return with new thinking and review the journey again.