Original · Unique Research · 2026-04-09
Editor's note: This is a historical roundtable reported by Unique Research on April 9, 2026. Company results, audience counts, prices, customer-acquisition costs, tool reliability and business predictions are retained as source or speaker claims, not independently audited findings. Haoshi Yinli, Yongbao Zhixu and MiaoSi AI are provisional English renderings; “Mr. Fu” is not identified more fully in the source. The narrative refers to annual fees of hundreds of US dollars, while the transcript describes a US$200 monthly plan paid annually; no annual total or discount is inferred. The valuation examples are hypothetical and have no stated currency. The source does not define the measure behind a “100,000-plus” report or “above 80 points.” Its introductory phrase literally reads “long-rental word of mouth,” an unclear expression; the later transcript discusses word of mouth without that qualifier. The quotation about doing passive things in radical times differs from the later quotation about making a passive retreat; both versions are preserved, and the attribution has not been independently checked. The genetic-data anecdote and its undefined “high-recall” description are reported business claims, not verified clinical performance or medical guidance. Dates and relative time references belong to the historical source.
Extraordinary Awards
If AI-Generated Content Keeps Getting Cheaper, Where Does Real Value Come From?
As Technological Advantages Are Rapidly Leveled, Scarcity Is Shifting from “Can You Make It?” to “Why Would Anyone Pay You?”
"
We all think we are incredibly capable. But back in the real world, how exactly can we use AI to make money?
Hundreds of highly skilled technologists work overtime to build an AI tool that supposedly generates anything in seconds, only to struggle with selling it. Meanwhile, a tiny two-person agency whose staff do not even write code wins business with AI and makes a fortune.
What explains the gap?
This is the cloud hanging over countless AI developers and content operators: if AI-generated content keeps getting cheaper, where does real value come from?
At an Extraordinary Awards Hangzhou AI WEEK trends roundtable Panel, MiaoSi AI initiator Mai Ge posed a searching question: “We all think we are incredibly capable. But back in the real world, how exactly can we use AI to make money?”
Around the table were Xiaohongshu marketing specialists, SaaS founders, and AI-platform growth hackers. These operators, getting results on the front line, pulled back part of AI's technological halo and opened the raw, real-world books on commercialization.
The Curse of Selling Fishing Rods: If Everything Is Transparent, Who Pays a Premium?
For many people entering AI, the first instinct is to build a tool—what is commonly called “selling fishing rods.”
But Xue Linfeng, a partner at 43 College, made a blunt statement: “When everyone is selling fishing rods, the rod will eventually sell at cost. That is inevitable.”
The article invokes a familiar economic idea: effective competition. When developers all call the same handful of foundation models and your capabilities are no more distinctive than your neighbor's, a company's net profit will ultimately approach zero.
Zhang Pinpin, founder of Yongbao Zhixu, felt this keenly. Working in customized ToB SaaS services, he had encountered a particularly difficult problem: how do you explain the bill to business customers?
AI depends on computing resources, and consumption flows like water—difficult to forecast precisely. Yet uncertainty is exactly what business owners dislike most: “They want the cost over a given period to become predictable and controllable.”
If you charge purely for compute, clients see a bottomless pit. Pinpin therefore chose an annual subscription based on estimated usage as a backstop. Even so, he remained clear-eyed: “Doing it faster and more cheaply can earn you a share of existing spending in the short term. But in the long run, there is destined to be no premium in that.”
Only one route remains: do it better. Not merely better technically, but by establishing what Xue Linfeng calls “a shared understanding of value.”
That shared understanding, and the premium, appear only when a client picks up your fishing rod and discovers that it not only catches fish, but catches bigger fish and more of them in a way suited to the client's own technique.
Magic and Standardization: When Prompting Becomes the Biggest Barrier
If selling tools is such a hard road, what about using AI to create hits yourself and build an entire content pipeline?
Shougong Chuan, founder of Lovstudio.ai, had a striking example: give AI a single initial prompt, have it write an article within one minute, publish it within five, and draw more than 50,000 reads across the internet.
Does that sound like a money-printing machine? Shougong Chuan also described the pain behind it.
The biggest problem was that it all looked too much like “magic.”
“There is a saying: the biggest barrier to Prompting is magic,” Shougong Chuan said. Without a ladder, you do not even know how to take the first step toward communicating with the best AI.
People think that all they need is a “super input box”: throw in a pile of files, and AI will produce a perfect result. But on a real production line, powerful workflows consist of layers of SOPs, feeding in only specific context at each step and continually adjusting the logic.
Turning that “magical SOP” into a standardized product that even a beginner can use—and getting customers to trust it and genuinely improve their output—involves an enormous gap in understanding. That gap is a chasm between AI and real money.
Chen Yina, head of growth at Happycapy, offered a different solution: decentralize and let the community help you build.
They built an underlying Agent operating-system platform and handed the stage to users. “Everyone's workflow is actually completely different.” Chen Yina found that when power users spontaneously built impressive automated text-and-image distribution workflows on the platform, sometimes feeding back into new features, their willingness to pay was remarkable. If you hit the right pain point, they would readily pay annual fees of hundreds of US dollars.
How Do You Sell It? First, See the Person.
One of the most painful things in the world is building a spectacular nuclear weapon, metaphorically speaking, and being unable to find a buyer to press the switch.
“Many exceptionally skilled people inside organizations can build all kinds of complex tools, but do not know how to sell them.” Wang Mengke, who has spent years managing Xiaohongshu operations for clients, pinpointed the weakness of technically oriented founders: “The difference is whether you can see the person.”
Business should begin with seeing a particular person's need and then making the product—not finishing the code behind closed doors and wandering the world with alien technology in search of someone to use it.
Wang Mengke's approach is blunt but effective: before the product has taken shape, publish a Xiaohongshu post to test the market. In a market where everyone is competing over existing demand, put the need out there first and see whether real people are willing to vote for it with real money.
Xue Linfeng went to another extreme: “anti-marketing.”
As paid acquisition becomes more expensive, spending tens or hundreds of yuan—or even more than a thousand—on a lead itself squeezes the room available for product value. Xue Linfeng chose not to spend a single cent on advertising. All orders were driven by what the introduction calls “long-rental word of mouth.”
It may sound a little arrogant, but the underlying logic was coherent: “That shared understanding is the most important marketing today.”
While everyone else throws their life into the traffic pool, returning to basics and relentlessly focusing on “business conversion and effective delivery” can put down the firmest roots in an increasingly restless system.
The Last Card for the Future: Preserve That Final Trace of Humanity
At the end of the conversation, Mai Ge asked: in this AI revolution, what kinds of people and companies will ultimately win? What will protect those that survive?
Wang Mengke's answer was down-to-earth: “Just staying alive is enough. Deliver at an honest price, and still be able to watch television after work—that is winning.”
Indeed, in a wave that can overturn everything at any moment, completing an OPC business loop and securing enough margin to keep going is already a great comfort to micro and small businesses.
Zhang Pinpin strongly agreed. His lesson came down to one word: rigor. “Release a new feature today, and tomorrow someone else can casually use a coding tool to copy it exactly. Where is the difference then? It is that clients know you can consistently deliver high quality, rather than being good one day and poor the next.”
For Mai Ge, data assets are what give large companies and operators confidence. “Can you carry strategy down into execution, then turn execution into a feedback cycle of data? That is the moat that large models will find hard to level.”
The most moving—and most painful—judgment of the session came from Xue Linfeng.
As the agent paradigm sweeps through everything, future large models may swallow all the gains from standardization. “The only thing that can ultimately let you win is retaining your human quality.”
He quoted a line circulating online: “Those who truly win do the most passive things in the most radical times.”
It was an almost romantic rejection of the rat race. When silicon-based minds can handle most work, the most valuable human ability will no longer be the number of lines of code written, or how quickly one produces a “100,000-plus” report. It will be the flashes of inspiration and emotional resonance that only humans can create.
Because that “human quality” is precisely the dialect AI can never learn.
More Details from the Conversation
Extraordinary Awards · Hangzhou AI WEEK Trends Roundtable Panel: “Content OPC: End-to-End Operations and Growth from AI Creation to Monetization”
Guests: Wang Mengke, Xiaohongshu Marketing Specialist at Haoshi Yinli | Zhang Pinpin, Founder and CEO of Yongbao Zhixu | Xue Linfeng, Partner at 43 College | Chen Yina (Yina), Head of Growth at happycapy | Shougong Chuan, Founder of Lovstudio.ai and AI KOL
Moderator: Mai Ge, Founding Partner of MiaoSi AI
Mai Ge: Many people may call themselves OPCs, but I do not think we are necessarily pure one-person companies. We all have our own companies and employees. Still, a very important question in AI right now is: how exactly can you monetize it? Monetization has a direct link to how you get an understanding of customers' or users' needs, and then satisfy them. So let us talk about something practical: how can we actually use AI to make money? My first question is whether each of you can briefly introduce your project, who your customers are, what pain point you solve for them, and how you charge.
Wang Mengke: I am Wang Mengke. I run a small new-media services company that helps businesses trade and sell products, mainly on Xiaohongshu. I am on the agency side.
Mai Ge: Xiaohongshu is a huge ecosystem. What exactly do you offer?
Wang Mengke: First, we sell our intellectual work. There are still information gaps in operating on Xiaohongshu, so we offer products such as courses, training, and ongoing hands-on coaching. We also sell our labor: we help clients find bloggers and influencers to say good things about them on the platform, in a planned way that is more focused on building a lasting body of content. Yes, that is what we do.
Zhang Pinpin: My business is fairly varied. The first part is customizing SaaS tools for companies. We currently work with industries such as publishing and healthcare, tailoring tools specifically for each business. The second is for companies that may not yet need their own custom tool, but do not use existing tools well. We provide AI services, helping them use the tools and delivering the results. For example, we help doctors produce science-education content for short-video platforms. That now has a fully digital, end-to-end AI generation and publishing process. Further down the line, we will develop ToC tools ourselves and make money through tool subscriptions or a share of compute charges. Those are roughly the different models.
Mai Ge: Understood. Would it be fair to say your tools lean toward AIGC?
Zhang Pinpin: Not necessarily, though some do. ToC is a major area. For example, in publishing, the tool replaces an editor's assistant: organizing verbatim transcripts, preparing initial drafts, and researching potential topics. It saves the work of one or two people across the publishing process, and I take an agreed proportion of the benefit from the part I replace.
Mai Ge: So, in effect, you are selling AI employees to help reduce staffing costs and improve efficiency?
Zhang Pinpin: Part of it works that way.
Mai Ge: I also heard you say you provide services, right? Some brands or clients may not be very good at human–machine collaboration themselves. Roughly what share of your revenue comes from products versus services?
Zhang Pinpin: In revenue terms, I currently think I will do less of both going forward. I will probably focus more on something platform-based for business customers. Even when the first two models generate steady revenue—for example, recurring maintenance revenue from custom enterprise SaaS—I do not regard them as good businesses, because they take so much of my time. I am not a pure OPC, but my company actually has only two employees.
Mai Ge: Understood. You still want a platform business.
Zhang Pinpin: Yes.
Xue Linfeng: The name 43 College probably gives everyone a sense of what we do. But fundamentally, I do not call it monetization; it is value delivery. As the old saying goes, teach people to fish. The first thing we deliver is the product itself. For AIGC content, that means delivering the AIGC output a company needs, whether text, video, or images—for marketing or publicity, for example. Then, at the next layer, what goes beyond teaching people to fish? Fishing rods. We can develop the rod, which, as we discussed in earlier roundtables, means building a tool. It may differ from a general-purpose tool on the market: it needs to be customized to your fishing technique and preferences. The next layer is teaching the art of fishing—how to fish—which is where College comes from. Beyond that, we teach you how to make a fishing rod that meets your own needs. Even if I develop one for you, it might not fit what you need, so I teach you to build the rod that lets you fish well yourself. That is a complete loop and the value we hope to deliver. I would not call it monetization; it is delivering value and receiving the corresponding value in return.
Mai Ge: May I ask a little more? What are your main paying users and customers like at present?
Xue Linfeng: They are mainly business customers. From our perspective, businesses are the ones that truly seek value. They want AI to bring their organizations value far beyond what a workforce composed of people could previously provide. So more of our customers are on the business side, including for training. Of course, training can also serve individual consumers, but demand from enterprises is actually stronger.
Chen Yina: I am Yina, head of growth at Happycapy. Happycapy essentially builds a platform that makes it easy for humans to communicate with Agents. How do we do that? First, we turn a cloud sandbox into an operating system for the Agent. It can then call different tools in that sandbox, operating its own computer just as we operate ours. On that foundation, the Agent and platform form a fairly general-purpose AI tool. It can help with most workflows; you can hand much of that work to AI. Comparable products include Claude and OpenClaw. Underneath, we also use Claude and MiniMax, but we add another AI layer on top so that you can orchestrate different models, together with a visual interface. That makes the platform useful to everyone from professional developers to complete beginners. Beginners can enter an out-of-the-box environment and use tools such as Claude or even MiniMax for automation. Our paying users currently focus mainly on development and content marketing, with automated marketing the most common workload. That includes generating images and video, as well as copy, documents, analysis, and Research in between. You can do it all in one place on the platform.
Shougong Chuan: I am Shougong Chuan, founder of Lovstudio.ai. Our main focus is fully automated content workflows. Let me start with a memorable case: we wrote an article in one minute, published it within five minutes, and received roughly 50,000-plus reads across the internet. Just one minute: give AI an initial prompt, and the article comes out. How did we make that happen? Of course, we have our own SOP. What I want to convey is that, in the AI era, content may become increasingly cheap and less valuable, much like code. This really is a transformative time. Within that, we may need to be guided by our own aesthetic judgment, our commitments to human qualities, or our company's mission and values. Those may be differentiators. Our services have two main parts. One is the fully automated content-creation workflow I just described, which can distribute across Xiaohongshu, WeChat, Twitter, and other platforms. The other is designing solutions around the Skills ecosystem for AI Agents.
Mai Ge: Listening to everyone, I feel most of the projects are still ToB, right? Only a few, such as Shougong Chuan's, seem more ToC. There is quite a lot of ToB here.
Shougong Chuan: Our enterprise solutions are ToB as well.
Mai Ge: So you also have enterprise solutions. Essentially everyone here has a ToB business. My own company is ToB too. We are called MiaoSi AI, and we help leading clients use AI to optimize the generation and placement of brand and performance advertising on Douyin and Xiaohongshu, to achieve better results. Since we all work in ToB, there is a question that has long bothered me and that I hope you can discuss. Most of our products and services are new and AI-based. They did not previously have established pricing systems, KPIs, or valuation frameworks. Now we need to sell them and set a price. What bottlenecks and core obstacles have you encountered in commercialization, and how have you addressed them? Mengke, your business is relatively mature. Why don't you start?
Wang Mengke: How do you sell it and set the price—is that the question?
Mai Ge: Yes. You may not struggle with this as much because Xiaohongshu training and advertising placement already have a well-established market. But I am especially curious about the others, whose companies make AI tools or services. Perhaps we should start with Pinpin instead. What is your commercialization bottleneck, and how do you solve it?
Zhang Pinpin: We are still talking about ToB, because ToC really is difficult. From the outset, we judged that we would struggle to collect money from consumers: my service does not help them earn money, and major companies offer things free on the consumer side, making it impossible to compete. One of the biggest early bottlenecks in ToB was that clients wanted costs over a given period to be predictable and controllable—for example, a fixed amount for the year. But SaaS tools consume compute, so usage has to be paid for. That problem held us up for quite a while. Eventually, the solution was something like a subscription: I estimate a usage level the client is unlikely to exceed, giving them a controllable expectation for budgeting and reporting. If they do exceed it, they still pay extra. That is roughly how it works.
Mai Ge: Do you charge a License fee for the product itself, then? Or do you charge purely for compute?
Zhang Pinpin: Currently, there is first a development fee for the specific use case. Then maintenance provides fixed, predictable revenue. For compute, I give clients an expected cost for a certain concurrency level or usage scale, which most can accept. Another model is payment entirely by deliverable. Take a medical use case such as interpreting genetic data. People began sequencing their genes more than a decade ago, because it was said this could estimate the likelihood of many diseases. That work used to require someone to spend roughly a week going through perhaps more than 1 million rows of genetic data, finding relevant information and evaluating it. With AI, that becomes much more convenient. But in healthcare, this is a very good high-ticket, “high-recall” offering. For that kind of work, I simply deliver one report and charge for it.
Mai Ge: Yes, that is a very good example. Linfeng may be similar: you mentioned delivering results as well as the tools behind them. In practice, is it easier to charge for results or for tools?
Xue Linfeng: Fundamentally, we are all building something called “a shared understanding of value.” Economics has a term, perfect competition—or effective competition. When the resources, content, and capabilities we can offer are identical—and the capabilities discussed in all today's roundtables can already address most enterprise scenarios—then under effective competition, your company's net profit will inevitably approach zero. The only thing we can pursue is a premium. Where does it come from? I think this is the biggest commercial bottleneck for everyone here: on top of a shared understanding of value, how do we make the customer feel that what we provide is more useful? You asked which part embodies our value. It is actually a complete loop: training is the starting point for establishing that shared understanding. When you use the fishing rod we ultimately deliver, you discover it differs from others. It can be customized so that you catch bigger fish and more of them than with someone else's rod. Only then will you recognize the value. That is where the premium comes from. When all of us sell fishing rods, the rod will eventually sell at cost. That is inevitable.
Mai Ge: Let me pursue that a little further. How do you make your fishing rod better to use than other people's?
Xue Linfeng: That goes deeper. From an AIGC perspective, the capability you provide includes what you, as a person performing above 80 points, can contribute—something we often discuss, for example, is taste. Does the fishing rod you make have better taste, storytelling ability, and architecture than someone else's? Everyone is talking about Skills today. Behind Skills is the current buzzword Prompt Engineer, right? How good your Prompt is determines how good the fish your rod ultimately catches will be. It may look as though AI can give you a perfect Prompt, but in reality it relies on the convergence of human experience and all that tacit knowledge. That is what our business values most: we consider our people our most important asset.
Mai Ge: So the Team actually developing these Agents is highly professional, with deep industry knowledge.
Xue Linfeng: Yes, with tacit knowledge.
Mai Ge: Understood. Your team consists of highly professional people putting their experience into the product. Yina, how do you sell your platform?
Chen Yina: We charge a monthly fee that includes a usage allowance. Because we are a ToC product and a general-purpose tool, users can essentially build their own workflows on our platform as long as they have a workflow in mind. What we need to build is a community. For us, monetization has roughly three parts: first distribution, second conversion, and third users paying. I think the first two are the hardest. But willingness to pay is very high. We launched a monthly plan at US$200, and users did not just pay US$200—they paid on an annual basis. So if AI can accomplish what they need their tools to do, they are actually very willing to pay. Most people in the SaaS market also know that many intermediate AI products, such as Claude and MiniMax, are not particularly reliable in calling tools to operate a computer. You therefore need Skills; later, we also introduced Agents so that you can build your own. Once it has memory, a Skill, and a process it needs to carry out, tool-call reliability becomes very high.
What is missing in the market is an easy way to look at other people's workflows and then build your own on a platform. Everyone's ideas and workflows are completely different. In our case, the commercialization bottleneck is distribution at the front end and finding a path that converts. There are simply too many AI products. Tell someone you have launched one, and before you say another word they are already thinking, “Oh, another AI product.” What should you do? You need to capture attention during distribution, in a very short time—perhaps two or three seconds of video copy—and show an advanced feature that resonates with them. We need to distribute the product in different formats on different channels. But once you reach the right customer, or a beginner is amazed by your Workflow, willingness to pay is very high.
Mai Ge: Put bluntly, everything ultimately comes down to selling, right? You still need to work out how to sell the thing.
Chen Yina: Exactly.
Shougong Chuan: One of our core commercialization bottlenecks is “workflow standardization.” I mentioned our Case earlier: we can write an article in one minute that reads very naturally and travels widely. But enabling clients to do the same has a real barrier. Have you heard the saying, “The biggest barrier to Prompting is magic”? Without a ladder, you get stuck at the first step: you do not know how to communicate with the best AI, let alone how to communicate better from there. Here is an example. Don't all AI products have what is called a “super input box,” where you can attach lots of files? GPT, Claude, Kimi—all these products are designed that way. But in daily practice, we find the best workflow is not to upload 100 files at once. It is to upload them one by one, explain what each file is for, and adjust your creative thinking step by step. So fundamentally there is a strict SOP. Handing that SOP to customers, or turning it into a product users will adopt and that genuinely improves their output, involves a substantial barrier.
Another issue is Context. AI is now very powerful, so everyone says, “Less control, more context.” But how do you give that Context to AI? Some people are also concerned about data security and do not want to give AI their data. These barriers create many commercialization bottlenecks. Every day, we work with people who are stuck at those points and battle through the problems with them.
Mai Ge: Next question. I also interview people in AI on my own account, and after speaking with many AI founders, I have noticed a major pain point: they do not really know how to sell their products. Many come from technical backgrounds and feel their product is excellent, but do not know how to sell it. Since we are discussing monetization, everyone here has clearly found some way to sell what they make. Let us be practical: how do you sell your products and services? Could you share?
Wang Mengke: I have seen that phenomenon often and feel it deeply. When I worked inside companies, many exceptionally skilled people could build all sorts of remarkable tools, but did not know how to sell them. I have many talented friends like that, and we discuss it frequently. I think the difference may be whether you can see the person. If we want to sell something, we should see the person first and then build, rather than build first and only then look for someone. That is the distinction.
Mai Ge: Analyze your users and their needs, and then find them, right?
Wang Mengke: Or, because we now have all these content platforms and AI tools, we can put Market before Build. We can test and promote the market first. Before my App has been built, I can publish a Xiaohongshu post.
Mai Ge: As an expert Xiaohongshu operator, you see the platform's traffic and lead generation as a good way to acquire customers, right?
Wang Mengke: Exactly. And you can validate it first. You no longer have to build behind closed doors. You can find out in advance whether anyone needs it and whether it solves a real person's problem. When you truly solve their problem, people will vote for you with real money, just as Yina described. That is certain; people are willing to pay.
Mai Ge: Finding your first customers on Xiaohongshu is a good approach too.
Wang Mengke: It is one strategy, one method that can save people time.
Zhang Pinpin: I strongly agree with Mengke: test first. I currently divide things into a few categories. First, I certainly will not build a product that does not make money unless it is an interest I would willingly spend time on even without selling it. Second, this is how I see AI's value: fundamentally, demand has not increased, so I look at where the money is. I think “being able to do more” is the least valuable thing. “Doing it faster and saving money” is the most basic level, and you can earn that money in the short term. Take a clear need: a company expanding overseas wants all its domestic social-media content translated into multiple languages. If I do that faster and more cheaply than another company, I can capture a share of existing spending. That works in the short run, but as we said, there will be no premium in it over the long term. After that, I can only “do it better,” because my understanding of the use case lets me do something other tools cannot. Only then can I earn a premium. The product really has to help someone earn more money or deliver better value before I can charge for that.
Xue Linfeng: Let me explain ours, which may be a little different. Although we need sales very much, our entire ToB operation is “anti-selling.” Since we started, we have not spent a single cent on new-media platforms or run any advertising. Fundamentally, marketing in the AI era continues the logic of earlier marketing: market insight, STP, user positioning, then integrated marketing communications, new-media pipelines, and so on. Follow that chain and you see what I meant by effective competition. Every cent spent on new media ultimately becomes a cost added to the value of your product. So we focus on only one thing. We have one North Star metric: customer conversion. We establish a shared understanding of value only through word of mouth. That may sound a little arrogant, but it is because we really do only one thing: value. We understand marketing very well, but I refuse to touch it. If I do, my attention shifts to making our offering sound more impressive and selling it to more people, leaving me less time to focus on value. It is somewhat counter-consensus, but all our orders really do come through word of mouth.
Mai Ge: Let me ask something that may run against human nature—or perhaps speak to it. Do you reward people who spread the word, or does it come entirely from their own genuine feelings?
Xue Linfeng: No rewards. People may understand the phrase I used a little differently from how I mean it: “a shared understanding of value.” I think that is enormously important in the AI era. Content will become more and more abundant, and reaching that shared understanding is extraordinarily difficult. Take Xiaohongshu or Douyin marketing: what does it cost today to bring a real customer into your own direct customer channels? Over the years, that went from a few yuan to tens of yuan, and eventually the average cost of acquiring a paying user rose to hundreds. In many lead-generation industries, it is now more than a thousand yuan per customer. That is what I mean by effective competition. This shared understanding is the most important marketing today, and it is what I focus on. We offer no rewards. Instead, we build very deep connections with users throughout the process. It may sound like boasting, but that is fundamentally what we do.
Chen Yina: We are a little similar. We do no paid advertising either, and we fully believe in and rely on product-led growth. But as a growth leader, you have to exercise considerable restraint. You need to recognize when something is really a product problem, without pushing product colleagues to tear the product apart every day, while still doing marketing and driving growth. For me, growth is an endless search for PMF. You have to keep searching across different terms and angles. After our first wave, we knew where that first audience was. Once you have found a distribution channel, can convert users, and raise their willingness to pay, you learn how much they will pay for that thing. Once the first angle is established, you can look for the next. The first also builds some trust and word of mouth. For example, only a month after we launched, power users in Chinese WeChat groups were sharing how they used the product and building tutorials and Skills for us. Overseas, one power user built a Reddit writing system that ranked first every day; another built a complete Instagram Caption workflow combining text and images, connecting everything from Research through distribution with one Skill. Some features had not even been officially announced: after power users built them, we asked whether we could turn their Skill into a feature in our product. Users' development drives both your product and your growth. The tricky part is that, initially, you cannot listen entirely to users. Before the product has taken shape, you must explore for yourself. But once you have a product, release quickly; then you can naturally see user feedback and adjust direction. That is the product-led growth path we have been exploring.
Shougong Chuan: I do not think it is so complicated. From a content-marketing perspective, there are only two elements: the person—the IP—and the content itself. So the question is whether your growth is IP-driven or content-driven. You can divide your company's business into four quadrants. A recent example is Mr. Fu: he has raised an OpenClaw with 30,000 followers. He is a high-value IP himself, and the content he shares is genuinely useful, so naturally it spreads widely. Many of us do not have such a strong personal IP. We may end up at the other extreme: some people mass-produce content with AI, but its quality is not high enough, so the word-of-mouth effect is weak. If I take content seriously and one piece takes off, it can trigger further chain reactions. Then you need to choose a platform. Xiaohongshu tends to amplify content more strongly, but the personal-IP effect is weaker—many posts receive tens of thousands of likes while the account has few followers. A WeChat Official Account may offer stronger stickiness. Personally, I lean toward IP-driven growth. I write in-depth articles on my Official Account to attract people interested in me, because those people may matter more to me, and then I try other forms of commercialization afterward.
Mai Ge: This has been a very enjoyable discussion. Because of time, let us move to the final question.
Wang Mengke: Mai Ge, aren't you going to share your own answer to the previous question?
Mai Ge: How do we sell our product? Our approach is fairly traditional. We define a profile for every client. Since we help brands optimize content advertising on Douyin and Xiaohongshu, we first identify the clients spending the most. They are likely to be very large consumer-goods or brand groups in categories such as beauty, mother-and-baby products, 3C consumer electronics, and home appliances. Most founders struggle with reaching those clients. The largest are undoubtedly the hardest to win because the existing interests and relationships run deep. Our next step is to find every possible way into the core decision-making circle. It is still a very traditional ToB acquisition approach, except that our engineering capability is strong. A single client may have a large budget, so we continue to approach this through efficiency. All right, the final topic: briefly, what do you think is a real barrier to competition in the AI era? What kinds of companies ultimately win? It will not be a world where everyone blossoms and each has a protected little patch. Linfeng mentioned the expert team behind a product, with strong Know-how. That looks like a barrier. Please share your thoughts briefly. Mengke?
Wang Mengke: What does winning mean? I think staying alive is enough. Survival is winning; that is my definition. I have only been running this small venture for a few years, and my feeling is that if you put care into every job, earn clients' trust, keep improving, and charge an honest price, you can stay alive.
Mai Ge: Understood. Steadily refine the product and create value, right?
Wang Mengke: If we do not insist on building a huge business, and instead want every colleague to do work they enjoy and still watch television after work—if we have come together by chance to work and live the life we all want, while providing value to clients—I consider that winning. If that is the definition, all you need to do is deliver well, build trust, keep learning and doing excellent work, charge reasonably, and sell to people who can pay.
Mai Ge: Understood. I think that is very good. It is a typical AI-era “one-person-style organization,” or OPC: people, machines, and AI finding their own point of happiness. That may be the state most people pursue in the future. All right, Pinpin.
Zhang Pinpin: I strongly agree with Mengke. This is actually my eighth year operating as an OPC, and after going through the pandemic, I think staying alive matters a great deal. We mentioned customer trust, but I see another layer behind it. One word is crucial: rigor. What does rigor mean? Being able to deliver a certain level of quality consistently and reliably. That is extremely important. In the past, if I built a new feature or tool, it might take someone else half a year to catch up. Now a programmer or product manager can use Claude Code to copy it exactly the next day. Where is my competitiveness then? Beyond drawing on tacit knowledge accumulated in the use case and doing the details better, the larger point is that clients trust me. The basis of that trust is that I can keep maintaining the standard, rather than succeeding sometimes and failing at others. I think rigor is very important.
Xue Linfeng: My conclusion comes down to one phrase: “human quality.” If you want to survive in the future, you must retain it. I also come from a technical background. In the foreseeable future, within not many years, the Agent paradigm will inevitably swallow everything. The one thing that can ultimately let you win is retaining that human quality. A line from the Bilibili creator Tang Zhi really struck me: “Those who truly win make the most passive retreat in the most radical times.” We already live in a very radical era, yet intangible cultural heritage has become particularly valuable. Retaining that trace of humanity may let you do things AI truly cannot do when models eventually swallow everything. It sounds pessimistic, but I think life may be quite happy when that day comes. Everyone can be a cog that is not alienated, doing something they enjoy. A genuine shared understanding of value can emerge. What we currently call a shared understanding of value is actually an alienated version driven by the economic system. If people can truly be human in the future, there will be more things they can do. We can be optimistic. Perhaps simply being alive will be a form of happiness.
Mai Ge: Your view is that, when material productive capacity is extremely advanced, everyone can have a rich and varied inner life, right?
Xue Linfeng: Yes, you could put it that abstractly.
Chen Yina: Human quality certainly matters for people. But for the product, our team follows the principle that speed is the one thing that cannot be defeated. I believe everyone has heard that in AI, speed wins. Yet doing it is not simple. In a marathon, in a world of information overload, how do you move that fast without becoming anxious, while retaining your humanity and still delivering high quality? That is difficult. A competitive barrier has to involve something difficult. But returning to the starting point, I think one such thing is speed.
Shougong Chuan: I strongly agree with Yina, though I may take it one level deeper. Most people know what is right, just as you can hear plenty of good advice and still struggle to live well. Suppose I ask you to go from a valuation of one hundred million this year to one hundred billion next year. You would consider that impossible. But if I relax the condition to one hundred million this year and 150 million next year, it becomes Achievable. You will find that the most important thing on that curve is your acceleration: establishing a positive strategy that lets your company, product, and market value keep growing under any circumstances. Faced with an uncertain era and opponents who do not follow familiar rules, we can shift toward a positive mindset and strategy, enjoy efficient growth, keep running faster, and ultimately reach the point where speed wins. I think that may be the core of surviving at the right pace.
Wang Mengke: What about you, Mai Ge? How do you win?
Mai Ge: What will large models level out? Some organizations previously had exceptionally dense talent or were very hardworking—advantages in management intensity. AI will level those out too. But I think two barriers can support success. First, an organization needs powerful data assets of its own, because the richness and accuracy of the data make a major difference to the strategies AI generates. Second, can it establish a closed data loop? I keep generating data, use it to generate strategy, move strategy into execution, feed execution back as data, and analyze how to correct it against the strategy, keeping the flywheel turning. Whether in Agent performance or strategy optimization, I think that may be a barrier. My impression is that some of the fastest-growing companies, such as ByteDance, may operate with a system like this.