Original · Unique Research · 2026-07-10
Editor's note: This is an interview-based profile of Bai Shuang, founder of Leapility, conducted on Wu Xiaobo Channel's "AI Future Talk" program. All statements, company names, and personal anecdotes are attributed to the interviewee. The 20 Q&A entries are preserved verbatim in spirit.
AI Industry Observation
How People Who Refuse to Become "Advanced Tools" Are Redefining Individual Value
From a translation student at ESIT Paris to an entrepreneur helping others productize individual value
During her years studying at the École Supérieure d'Interprètes et de Traducteurs (ESIT) in Paris, Bai Shuang did this many times: translating a viewpoint she didn't personally agree with, word for word, from French into Chinese. She knew the principles of fidelity, expressiveness, and elegance, but the act of translation itself defined her position—you can only polish, you can't change the original meaning. Once after translating, a thought arose in her: "I seem to have become an advanced tool."
This thought later determined the trajectory of her entire later life. After undergraduate graduation, she was the only Chinese student in her cohort; her French-Chinese-English language combination was rare, and the opportunity to do simultaneous interpretation at the UN was on the table. Her teachers felt it was a shame, but she was certain: she wouldn't do it.
"I love languages because I want to use them as a medium to open up the culture behind them. That excites me—not being an advanced translation tool."
This past story came up when Teacher Wu Wei and Bai Shuang chatted on Wu Xiaobo Channel's "AI Future Talk." The conversation's theme was "how ordinary people's experience can be monetized in the AI era," spanning nearly two hours, covering her entire journey from Paris to London, from education technology to two rounds of entrepreneurship. But if you trace a single thread through this conversation, you'll find that what truly matters isn't the word "monetization," but the same question Bai Shuang has spent over a decade repeatedly answering: what should a person do to not be turned into a tool.
This question had already been brewing in her for years before she switched to the Agent track in late 2022. For her first startup, she did AI translation, raising one funding round each in early and late 2022, totaling 10 million RMB. After ChatGPT appeared, she quickly realized she had to pivot—not just because the technology changed, but because she'd heard too many top investors tell her the translation track was too small, and the human-AI collaboration model didn't have enough scalability. Her later explanation was blunt: "The business model we talked about was essentially still me selling my own service, just with higher efficiency." This sentence is almost the judgment standard that recurred throughout her entire interview.
What's Beneath the Iceberg
Bai Shuang's answer to "will experience be eaten by large models" bypasses the two answers most people habitually give—not "experience is still important," nor "experience will be eliminated," but first makes a distinction: past experience and incremental experience are two different things.
She used an iceberg metaphor. Large models have already learned the part of the iceberg above water—that's the explicit knowledge accumulated over human history, constituting the model's capability boundary. But beneath the iceberg there's a vast amount of tacit knowledge that hasn't been written down, hasn't been indexed by the internet.
More importantly, the iceberg itself is also changing. New business scenarios, new user behaviors, new marketing methods continuously produce new experience. Bai Shuang gave the Douyin example: before Douyin appeared, many people couldn't have imagined "live commerce" would become a mainstream marketing method; before that, there was no mature set of live-commerce experience in the industry until someone first entered the field, encountered change, and then consolidated methodology.
This judgment directly leads to a concept she calls "Token + X." The money in the Token part will be taken by model vendors and compute companies; most people in this track are payers—using Claude, using Codex, essentially paying for general-purpose capabilities. But there's also a long-tail market she calls the expert agent track—this money won't flow to model vendors, but to people who continuously acquire fresh knowledge in vertical domains and are trusted by others. X is this expert premium independent of general-purpose models.
This judgment sounds a bit like the old saying "what others don't have, I do," but Bai Shuang told a specific, almost counter-intuitive example that landed this theory. Their company has a post-00s employee who studied film before, and after joining was assigned to overseas Reddit operations, initially not understanding how this space worked at all. Three months later, she figured out a method that could stably post, avoid bans, and bring continuous traffic. She organized this method into a Skill methodology, posted it on Xiaohongshu, saying she'd made an Agent and inviting people to use it. Overnight, 200+ people subscribed. Bai Shuang repeatedly emphasized that this girl wasn't some "old master." Behind this sentence is actually her correcting a widespread misconception—many people think the expert agent track naturally belongs to people who've soaked in an industry for ten or twenty years. Her judgment is exactly the opposite: what's truly scarce isn't years of experience, but a person's ability to quickly learn in an unfamiliar domain and consolidate what they learn.
The Exposed "Knowledge Base" Demand
If Token+X is the truth Bai Shuang has figured out in recent years, then the pit she stepped on in her second startup was a costly validation. After pivoting from the translation project, she saw that US companies like Glean doing enterprise knowledge retrieval were doing well, judged there should be a corresponding opportunity domestically, and went to research a batch of enterprises wanting to build knowledge bases. The conclusion from the research was a sentence that even she found piercing: "The needs users give you might be fake needs." Many enterprises say they want a knowledge base, not necessarily because they truly need it, but because of top-down anxiety—leadership feels they should have an AI, so they package this anxiety into a procurement requirement. Even if such a system is actually deployed, it's hard to measure productivity improvement, ROI doesn't add up, and private deployment costs are high. In the end, Bai Shuang's judgment was: this path, she can't do and doesn't want to do.
She extended this lesson to a larger judgment: developer-facing tool products like Dify essentially solve "who writes the code," not "who understands the business." A Fortune 500 company's IT department may have strong technical capability, but what points to review when a clothing company purchases fabric, or exactly which process环节 the legal department gets stuck on—IT has never truly understood. Over the past two years, many workflow products have entered enterprises but failed to deploy; the root cause is the same—developers provide capability, but business know-how is something completely different, and the gap between the two is something AI currently can't fill. This is also why she later judged that in this wave of technology, the real wealth redistribution won't benefit developers, but those who truly hold industry know-how and are willing to continuously deepen.
The Term "OPC" Is Being Used Wrong
The part of this interview where Bai Shuang showed the most emotional fluctuation was correcting the concept of "OPC" (One Person Company). She gave an example: a developer who used to take outsourcing projects with a two-month delivery cycle, now with AI it's shortened to two weeks, efficiency multiplied—is this person an OPC? Her answer is no, because the delivery主体 is still himself; the moment clients go from 1 to 100, he still can't stack time and can't scale. This is essentially still selling time, just with higher unit efficiency.
Her definition is "solo but scalable"—one person, but scalable. The true OPC in the AI era isn't about "one person doing many people's jobs," but "can you productize individual value"—selling a product, not time.
Under this definition, she proposed a formula for measuring individual value called AIM: A is Agent, corresponding to a person's professional depth, determining whether the Agent they build can outperform a general-purpose Agent; I is Influence—a person with only capability but no influence is like a company with R&D but no marketing; M is Marketability, market power, which needs to be補ed up by platforms—building, distribution, transactions, payments, these commercial infrastructure pieces shouldn't need an individual expert to build from scratch.
She holds a near-pessimistic attitude toward pure Skill marketplaces. She mentioned that Skill Hub has already become a bit of a junkyard—hundreds of thousands of Skills listed, most of them garbage, for a simple reason: most people contributing these Skills are developers, not people with real know-how. She used Shopify as an analogy: Shopify was never a shelf; it provided user profiles, CRM, copywriting generation—a whole set of infrastructure for business owners. She believes expert individuals need the same thing, not a priced, download-and-leave Skill store—in that model creators' assets have no protection, and users can even resell after downloading.
Swarm Intelligence, and People Who Aren't Experts But Can Participate in the Ecosystem
Regarding future organizational forms, Bai Shuang cited KK's observation in Out of Control—individuals first achieve "super individuation," then combine into a "swarm," with Hollywood-style project-based collaboration as the prototype: making a film, bringing the top people together temporarily, disbanding after filming, reassembling for the next one. Her judgment is that pure brain-labor professional-service companies with no physical deliverables—especially consulting, which naturally "follows the person, not the company"—will move toward this form first; while enterprises producing physical goods like phones, food, and cosmetics won't disappear as quickly. Her own company now has 12 people, no duplicated roles, each person commanding their own cluster of Agents, then collaborating with other colleagues' Agents—this is already a swarm inside the company.
But she specifically reminded one thing: not everyone needs to, or can, become an expert themselves. The ecosystem also has many roles that don't directly produce professional content but can participate in value distribution. She designed three types of agents—Sourcing agents who find real experts in a vertical domain, coaching agents who help "old-master-type" experts turn their experience into one-on-one Agents, and IP agents who help people who've already built assets amplify their reach. These three roles may not be experts themselves, but can participate in the ecosystem's收益 distribution through coaching, incubation, and revenue sharing. The judgment behind this design is: people who weren't content creators in the last internet cycle still have a chance in this one.
Sleeping Beauty Demand
Toward the end, the principles Bai Shuang talked about are actually farthest from technology. She said entrepreneurs shouldn't only stare at "my product, my technology," but should think more about human nature—Pop Mart's barrier isn't in its supply chain, but in its precise embedding into the human desire to open blind boxes; this demand existed decades ago, just in a different form. She calls these not-yet-fully-satisfied, or even not-yet-recognized, demands "Sleeping Beauty demands." She mentioned that her grandparents' generation's later years were actually quite boring, and young people are lonely too—two-thirds of her company's 12 people have pets, the need for companionship has never disappeared, it just hasn't been taken seriously. Behind this passage is her most direct answer to "where should AI entrepreneurship look": technology barriers are being rapidly leveled, engineering capabilities are increasingly easily补ed by AI, and what's truly hard to replicate is acute judgment of human needs.
Leap
Near the end of the interview, Teacher Wu asked what she'd given up over these years of entrepreneurship. She said she gave up a lot, but didn't feel it was sacrifice, because they were all actively chosen, and what she's doing now happens to align with her passion. This is probably also why she named her company Leapility—Leap your ability, which she translates into Chinese as "跃向" (Yuexiang). From the translation student in ESIT Paris who felt she was an "advanced tool," to today's entrepreneur helping others productize individual value, Bai Shuang has walked this path for nearly fifteen years, switching tracks twice in between, giving up the UN simultaneous interpretation opportunity. But what runs through it all is actually the same thing: no matter how times change, no matter how much AI takes over tool-based work, the question a person ultimately must answer is the same—have you found the thing you don't want to be replaced from?
Interview Selected Q&A
Q1｜You studied linguistic science in France. How did this experience influence your later AI Agent work?
Bai Shuang: It influenced me quite a bit. I went to France right after high school; my initial goal was very clear: I wanted to do simultaneous interpretation at the UN. Because French is a UN official language, and I'd studied English as a child, so my thought was I'd learn Chinese-English-French and become a UN simultaneous interpreter. Half a year after arriving in France, I entered Paris Descartes University to study linguistic science. Linguistic science, put simply, is studying how the human brain uses natural language. Later when I did Agents, this learning experience gave me a lot of first-principles support. This also explains why in June 2023, when people weren't really talking about agents yet, I'd already caught this direction. Because I look at large model breakthroughs from the human brain's language centers—the listening, speaking, reading, writing centers and cerebral cortex. I felt at the time that technology had matured enough to simulate human behavioral patterns across perception, thinking, action, and memory, to have it do some work.
Q2｜You clearly had the opportunity to do UN simultaneous interpretation—why did you ultimately give it up?
Bai Shuang: After I entered ESIT, I started receiving very orthodox translation training. But it was also during this process that I discovered my love for languages wasn't about becoming a translation tool itself. Because when you translate, you go from language A to language B, and can only polish in between. We talk about "fidelity, expressiveness, elegance," but you can't change the original meaning. But sometimes we'd translate things that I personally didn't agree with, and I still had to translate them out. In that moment I'd feel like I'd become an advanced tool, which made me very frustrated. I later realized I love languages because I want to use them as a medium to open up the culture behind them. That excites me—not being an advanced translation tool. So after undergraduate, I didn't continue to graduate school. Actually I was the only Chinese student in my cohort, and my language combination was very rare, so the UN opportunity was significant. My teacher felt it was a pity, but I was very certain: I wouldn't do translation anymore.
Q3｜Your first startup was still AI translation—why did you later choose to pivot?
Bai Shuang: Because I was very clear at the time that the translation track wasn't a particularly big one. My first startup project was AI and translation-related; we got one funding round each in early and late 2022, totaling about 10 million RMB. Then ChatGPT came. At the time I thought I must combine new technology and pivot. Because when doing the translation project, I'd talked to many top investors, including partner-level ones. They all thought the translation track was too small. Plus our model was human-machine collaboration, and scalability wasn't that strong. So we later pivoted the translation project to the Agent track. At first we also did vertical Agents related to translation, and later it became today's creator platform for super-individuals and expert individuals. I'm very glad I'd read The Lean Startup before starting a company; it has an important concept called Pivot. So I don't think the first startup direction switch was a failure. I think in a rapidly changing era, pivoting is the normal thing.
Q4｜In the large model era, does ordinary people's experience still matter?
Bai Shuang: I think first we need to distinguish: we're not talking about past experience, but incremental experience. Now is a rapidly changing era; past experience might not serve tomorrow's needs. What's truly valuable is experience that hasn't happened yet but needs to be acquired by someone. This part actually requires a person as a subject interacting with the world to continuously acquire, think, and consolidate. If we compare knowledge to an iceberg, large models have already trained on the explicit knowledge above the iceberg, forming their capability boundary. But beneath the iceberg there's still a lot of possibility, like personal, expert tacit knowledge. In the future this tacit knowledge will also be slowly dug out, and even part of it will be internalized by models. But the world keeps changing, and new iceberg bottoms will continuously appear. For example, the way marketing is done in the future might not look like it does today. Before Douyin appeared, it was hard to imagine "live commerce" would become a mainstream marketing method. So experience isn't valueless; it just belongs to people who keep exploring vertical domains and continuously entering the field.
Q5｜Why do you say "general AI provides general services, professional agents provide professional services"?
Bai Shuang: When I interviewed KK last year, he made a point: general agents are like Swiss Army knives. A Swiss Army knife can do anything, can peel apples, can do many things, but you can't use a Swiss Army knife to perform surgery. In this world, we need both general services and professional services. This won't change. Only the future delivery method will become: general AI provides general services, professional agents provide professional services. These two tracks are completely different commercial landscapes. General agents are centralized; they'll be owned by model vendors in the future. The model is the general Agent. So startups doing general Agents today but not owning a model—I think they're all in danger.
Q6｜So where is the ordinary person's opportunity?
Bai Shuang: The ordinary person's track is the second market, the expert agent track. The first market is centralized and model-driven. This money will indeed be collected by model vendors and compute companies. Most of us in this track are payers. For example, using Claude, Codex is paying for general models. But the second market is very long-tail. Expert agents require people to acquire fresh knowledge and continuously iterate. This is something model attributes themselves can't do. If you can provide some of humanity's fresh knowledge ahead of large models and form your own community, it can work. But this must be trust-driven. Why would others use your expert Agent? Because they trust you, because you have influence, have your own private-domain community.
Q7｜What does your "Token + X" mean?
Bai Shuang: Token is the money large model vendors earn, and X is the expert premium independent of general models. What users are truly willing to pay for isn't just model call cost, but the judgment, methodology, and trust that you—this person, this expert, this professional knowledge system—bring. I think it's hard to support this kind of transaction solely with a Skill marketplace. With so many Skills on there, how do users know whose to use? In the end it's because they trust you that they buy. So if you can combine your own know-how, influence, and private-domain community, users aren't just paying Token fees—they're paying the expert premium. This X is the part ordinary people can monetize.
Q8｜Must experts be "old masters"?
Bai Shuang: Not necessarily. Let me give an example of someone on our team. She's post-00s, studied film before. After joining the company, we had her do overseas Reddit operations. When she first came, she barely knew how Reddit operations worked. But after three months, she'd accumulated a set of experience that was validated: able to post stably, not get banned, and bring continuous traffic to our website. So she made this into a Skill methodology, posted it on Xiaohongshu, saying she'd made an Agent and inviting people to use it. Overnight 200+ people subscribed, and she quickly had her own private domain. Do you call her an old master? No. She's someone who can quickly learn in new domains. So some experts are indeed old masters, but more importantly, do you have curiosity, do you have the ability to quickly learn, validate, and consolidate methodology in new domains?
Q9｜How can ordinary people turn their experience into reusable assets?
Bai Shuang: We think the interaction method should be natural language, even voice conversation. But the problem is, many people "don't know what they know." So we need an Agent to guide them. There used to be a job called consultant, or knowledge extraction. What we're doing now is using an Agent to complete this. On our platform, this role is called Creator Helper. It first helps you do commercial value positioning. Just like an entrepreneur shouldn't start with "what technology do I have, I'm holding a hammer looking for nails," but should first ask: what is my value as a person? You first need to do digital twin commercial positioning. Then based on the value points you provide, converge them into Skill asset packages. After that, how do these Skills connect to help users complete their journey? Users won't say "Mr. Wu, can you use one of your skills to help me." Users will only say: "Mr. Wu, can you help me get this done?" So there's a lot of productization work here, not just staying at the "build Skill" dimension.
Q10｜Why do many Skill markets become "junkyards"?
Bai Shuang: I think many Skill Hubs have already become a bit of a junkyard. Hundreds of thousands of Skills, but most are garbage. Why? Because most people contributing these Skills are developers. I might sound a bit offensive, but I think the wealth redistribution from this wave isn't going to developers, but to people who truly have know-how. For example, can a Fortune 500 company's IT make a particularly useful business tool for the legal department? Or what points does a clothing company review when purchasing fabric? IT can't understand easily. Because as a developer, you only have coding capability, not business know-how. So what you make might not be deployable at all. This is also why many workflow products have been hard to deploy in enterprises over the past two years. My judgment is that those truly benefiting from this wave are people with curiosity who keep deep-diving into vertical domains.
Q11｜Why did you pivot from To B enterprise knowledge bases to expert individuals?
Bai Shuang: To B is of course hard to do. At the time we built knowledge bases based on RAG. Back then Agent form wasn't particularly clear yet; enterprise search, enterprise knowledge base looked like a relatively mature direction. But we quickly learned a lesson: the needs users give you might be fake needs. Users saying they want a knowledge base might just be a need driven by top-down anxiety. They might not truly need it, and might not be willing to pay for it. If after deployment it can't bring high productivity improvement to the enterprise, ROI doesn't add up, then in the end it's just a cold Q&A bot. You can't deliver results just by "asking." And private deployment is expensive. So after going down that path, I felt it had no value.
Q12｜Why do you keep emphasizing "deliver results," not "deliver tools"?
Bai Shuang: My first translation project actually took off because I didn't sell software to clients; I was delivering results. Foreign companies already had hundreds of thousands in annual translation budgets; they used to go through translation companies. What I did was a very certain track replacement: same results, I used technology and process restructuring to bring costs down. Clients don't care what tools you use behind the scenes; they only look at results. If you try to sell software to a foreign company, the cycle is very long—security, private deployment, procurement process all complex. But if you deliver results, you can quickly build cash flow. So when I did my second startup, I was also very insistent on this: I don't want to deliver a tool; I want to deliver the final result to you.
Q13｜Why will natural language become the core of next-generation product interaction?
Bai Shuang: Because language itself is the carrier of information. I study languages, so I very firmly believe one thing: everything that can be said can be consolidated in natural language. When you train an intern, you find that basically everything you teach them is expressed in natural language. So if the Agent is natural-language-driven, then natural language can definitely carry a considerable amount of information. Future products shouldn't make users learn "click this, click that." The problem with much of the old software is that you had to first learn how to use the software before you could get things done. But this is actually counter to human nature. The more natural way is: you directly say what you want to accomplish, the system understands your goal, then helps you get results.
Q14｜What do you think about many people being obsessed with Vibe Coding and rapidly making products?
Bai Shuang: I'm not saying being obsessed with Vibe Coding is bad, but I think before entering this state, everyone should first think clearly: what value does this thing you're doing have for society? Everyone can slow down a bit and think about who you're doing this for, what problem you're solving. This world definitely has many "Sleeping Beauty demands" that aren't met. But we shouldn't think of something and immediately go do it. You need more thinking: is this thing truly solving real value for a certain group? I think the problem with many entrepreneurs now is that the thinking period at the front is too short. Especially for OPC, many people think "because I can code, I see a need, so I immediately go do it." But in fact, you need to first research what the commercial value actually is. This is entrepreneur thinking, not developer thinking.
Q15｜You say "see the path and don't take it"—what does this mean?
Bai Shuang: This is a viewpoint from the novel Sky of the Curtain. The paths others have walked were valid under the conditions of their time and place, but that doesn't mean they're still valid now. Much experience has spatiotemporal conditions. So experience of course has value, but you can't blindly worship it. You need Critical Thinking to judge whether this experience still holds today. I think the AI era especially needs this ability. Because change is too fast, you can't just copy the paths others have walked. You need to understand the first principles behind it, then find your own ecological niche.
Q16｜What is a true OPC? Is one person using AI to increase efficiency called OPC?
Bai Shuang: I especially want to say that many people's definition of OPC is wrong. Simple example: if you used to do development for others and delivering one project took two months, now with AI it's two weeks—are you still selling your time? Just unit efficiency increased. If the delivery subject is still you, you can't scale from one client to a hundred. With more clients, you still have to stack people, because human time is always limited. So true OPC isn't higher efficiency, it's Solo but Scalable. You need to productize individual value. What you sell isn't time, it's a product. This product can be replicated, can serve more users, rather than each additional client consuming another chunk of your time.
Q17｜If everyone becomes OPC, what will future organizations look like?
Bai Shuang: I think the future will first super-individualize, then swarm-ize. KK talked about "swarm organizations" in Out of Control—multi-center but hyper-connected. Many of our companies and platforms now are centralized, but in the future a lot of brain labor will become another form. You can understand it as the Hollywood model. When Hollywood makes a film, it organizes a group of top talents to team up temporarily. After this film, everyone disperses; for the next film, they reassemble. Many professional services will be like this in the future. End-user needs haven't changed—they still need consulting, research, professional services. But the way needs are satisfied may become a combination of multiple OPCs. So the needs themselves haven't changed; what's changed is how they're satisfied.
Q18｜Not everyone can be an expert—can ordinary people still participate in this ecosystem?
Bai Shuang: No. We designed three types of agent roles. The first is the Sourcing agent. They identify who are real experts in the industry, a bit like headhunters, or what expert network companies do. The second is the coaching agent. They coach experts, especially some older experts, helping them extract experience and turn know-how into Agents. The third is the IP agent. They help experts amplify traffic and commercialize. So not everyone must become an expert. Some are suited to be experts, some to identify experts, some to help experts productize, and some to help experts with distribution and monetization. A three-party revenue-sharing mechanism will form here.
Q19｜AI product features are easily copied, and models are iterating fast. So what is an AI startup's moat?
Bai Shuang: Product is just a threshold, engineering capability is also a threshold. I think the AI era raises higher requirements for entrepreneurs: commercialization thinking, strategic positioning, and long-term thinking become more important. You first need to think clearly about your ecological niche relationship with model vendors. Will what you do be eaten by model vendors? If a model company does this, will users worry their assets are being distilled? If so, then neutral companies without models actually have an advantage. Second, you need to think about whether over time you can consolidate your own moat. If your users come just because one feature is cheaper or better, then when someone has a cheaper, stronger feature, users leave. What truly matters is whether you can consolidate things beyond products—like service ecosystem, expert assets, industry resources, and trust relationships.
Q20｜If you could give one piece of advice to ordinary people wanting to try OPC or find opportunity in the AI era, what would it be?
Bai Shuang: Don't put labels on yourself first. Don't say I'm a liberal arts student so I don't understand technology; don't say I'm an ordinary person so I have no opportunity. In the AI era, what truly matters is whether you have curiosity, the ability to keep learning, and taste. I think AI taking away our tool attributes actually gives us the chance to return to creativity. In the past many people were alienated into tools by 996. When a person's tool attribute is amplified very large, they become numb and lose creativity. But if AI can help you do a lot of tool-based work, even bring you some passive income, then your time and money freedom truly appear. At this point, what should people do? Return to your passion, keep exploring new things. I always say find your Spark. What you truly love might be where you reconnect with the world. The universe rewards those who firmly choose passion. Leapility in Chinese is "跃向," meaning Leap your ability, an ability leap. We're not saviors; we're just proposing a stance: use AI to achieve value leap, to choose a brand new way of life.