--- title: "Former DingTalk VP Starts Up: Employees Code All Internal Systems Themselves" author: "Unique Research" sourcePublication: "Unique Research Substack" originalPublishedAt: "2026-05-16T12:09:22+00:00" canonical: "https://ffcap.cn/en/research/src-20260516-01html" source: "https://uniqueresearch.substack.com/p/src-20260516-01html" language: "en" --- # Former DingTalk VP Starts Up: Employees Code All Internal Systems Themselves _Original · Unique Research · 2026-05-16_ _Editor's note: The first-person report and its judgments belong to the original Chinese author. This English rendition retains the opening essay, the full interview narrative, and the complete selected Q&A. Career history, financing, company scale, Token consumption, and business-model figures are source or speaker claims, not independently audited findings. Company names (K2 Lab / 攀峰智能, Moras) and personal names (Wang Ming / 王铭) are working English renderings where official English forms remain unverified. Platform names (DingTalk, TikTok, WhatsApp) are retained in common English usage. The term "Code" in the title and throughout is retained as the source's English loanword for programming. Monetary figures are preserved as stated in the original (US dollars where specified). The source is dated May 16, 2026._ AI Future Talk "Life doesn't necessarily have meaning, but you've got to do something" " He had just resigned from his position as VP at Alibaba DingTalk, registered a company called K2 Lab (Panfeng Intelligence), built an AI Agent that helps TikTok creators make money—and doesn't charge subscription fees, but takes a cut based on performance. This is his fourth startup. K2 is the code name for K2, the world's second-highest mountain at 8,611 meters above sea level. But in mountaineering circles, its reputation in some ways surpasses Everest—because its climbing fatality rate is close to 25%, ranking first in difficulty. I chatted with Wang Ming for nearly two hours on the Wu Xiaobo Channel's "AI Future Talk" live stream. Can't Stay at Big Tech Anymore Wang Ming's resume follows a typical serial-entrepreneur path. He started a business right after graduating college in 2009, building a delivery platform similar to today's local-life services—the model was too early, mobile internet hadn't arrived yet. He sold the company in 2012 and joined 58.com. There he worked on e-commerce-ifying life services, and later internally incubated 58 Daojia, which reached a Series A of US$300 million in one to two years, with a valuation of US$1 billion. Later he went independent again and built a SaaS plus B2B trading platform. Before 2019 he had traveled across seven continents and four oceans. During the pandemic in 2020, he considered starting a rocket-building company—he genuinely went around talking to investors and practitioners, but ultimately decided it was still a regulated industry and the timing wasn't right. Then he joined Alibaba DingTalk, first working on SaaS ecosystem building, and later taking charge of the large-model ecosystem and AI-native products. During his years at DingTalk, he did one thing: dealt with hundreds of AI startup companies. Helping Alibaba build its large-model ecosystem, examining these companies' products, technical routes, and business models. After seeing enough, he couldn't sit still. "Startup companies are so fortunate. In this era of rapid change, they can spend 100% of their energy learning and creating." He said, "But at a big company, because the organization is large, you have to do a lot of management actions to fight entropy increase, and a lot of time is consumed internally." He also used a metaphor: "A big company has Yao Ming's height—it can dunk on tiptoes, but it can't be as agile as a 1.7-meter point guard." This isn't about big-company disease. What he means is that size and agility are inherently contradictory—it's a law of physics, not anyone's fault. But in the face of change at AI's speed, agility may matter more than height. "The evolutionary speed of this era is unimaginably fast, and the decisive battle for the final ecological niche will come earlier." At the end of last year, he ran out. Not SaaS, but "Helping People Make Money" Many people would find it strange: someone who spent years building SaaS ecosystem at DingTalk, leaves and doesn't build enterprise tools? Wang Ming's explanation is very direct. During his years at DingTalk, he came into contact with a large number of Chinese To B companies. He arrived at a judgment: in the Chinese market, everyone treats AI as a tool, and as a tool it's hard to command a high premium. Users think, "I might as well develop one myself." Plus his own company is a living example—K2 Lab was founded a few months ago, already has nearly 40 people, and the HR system, operations system, and financial BI are all coded by employees themselves using Claude Code. "Monthly Token consumption is roughly equivalent to 30,000 employee salaries." I pressed further: so isn't this even worse for SaaS companies? He smiled and didn't deny it. So K2 Lab chose a different path. Their product is called Moras, an AI Agent for the TikTok ecosystem that helps overseas creators do commerce. The keyword is "helping creators make money," not "providing tools for creators." What's the difference? The business model is different. It's not selling subscriptions or collecting monthly fees, but paying based on performance—helping creators make money and taking a cut from it. Wang Ming has thought about this thoroughly. He divides product value into three layers: helping people make money, helping people save money, and providing emotional value. In his view, the "making money" direction has the greatest potential. "Saving money is nothing more than replacing labor, and large-scale labor replacement isn't that easy—it doesn't quite fit the laws of development. But creating value and making money directly, people embrace it very readily." "AI Hires People" When talking about Moras's specific product logic, Wang Ming used a concept that left a deep impression on me: AI hires people. The traditional human-machine collaboration model is "people use AI"—people lead, AI assists. But when building the product, they found that overseas users are quite lazy. Most AI capabilities are already hidden behind the scenes, the interaction is made very simple, and users still find interacting with AI troublesome. Someone directly said, "Why don't you just hire me?" This sentence inspired them. In Moras's logic, AI is the leading party. Product selection analysis, content scripts, data review—theseaspect AI runs on its own. But someaspect are too costly for AI to handle. For example, identifying hallucinations in content, judging whether a selected product is truly suitable for a specific audience. These are things a human can tell at a glance, but AI consumes massive compute. So they let users do these "gatekeeping" tasks. Users who are willing to do them get a higher cut; those who aren't willing, letting AI brute-force compute, get less money. "Even the best autonomous driving isn't as good as the worst human driver," he said. "People are inherently better than AI in certain scenarios." This logic flips the employment relationship: it's not people hiring AI to do work, but AI hiring people to do what AI can't do well. What people get is a "base salary," and the level of the base salary depends on how much judgment you're willing to contribute. I find this思路 quite interesting. It hints at a possible future form: agents don't replace people, but become a kind of "employer," and people become a node in the agent network, responsible for handlingaspect where AI isn't economical. AI Native Is Global Native Why go overseas? Why not stay in the domestic market? Wang Ming has probably been asked this question many times, and his answer is crisp. "In the past, when doing software going overseas, every country had different management methods and habits, and you had to redo the interface. But AI Native is approximately equal to Global Native." His logic is: if your product directly delivers results (helping creators make money), then users don't care what the interface looks like or what language it's in. What every country wants is the same result—making more money. Crossing the barriers of language and interface, the company is inherently global. And what they connect is the Chinese supply chain. Over 50% of products sold through TikTok commerce are Made in China. Taking Chinese supply-chain capability overseas is itself a huge structural opportunity. In terms of具体implement, their first wave of seed users was ground out one by one through cold email and WhatsApp. "Users don't care at all whether your phone number is +86; communication is very smooth." For market selection, they directly chose the United States. They didn't choose Southeast Asia, and the reason is also very practical: compute costs are expensive, Southeast Asian labor is cheap, the math doesn't work out. If hiring local people is cheaper than running AI, the business logic of an AI Agent doesn't hold. SaaS Subscription Fees May Disappear When talking about changes in business models, Wang Ming's judgment is relatively radical. He believes the agent era will turn SaaS's subscription model from mainstream to transitional. The logic goes like this: in the past, SaaS charged subscription fees because software's marginal cost was extremely low—just run the servers. But the AI era is different; every day consumes compute, buys Tokens. If you only provide a tool that makes others comfortable to use, without directly creating result value, your added value is too low. "In the future, not only will there be performance-based payment, but the traditional advertising model may also disappear. When you help people deliver results and take a cut from it, the proportion of added value will be very high." This differs somewhat from the judgment of QuickCEP's Billy, whom I chatted with last week. Billy feels that pure performance-based payment is unfair to the service provider, and the intermediate state is compute-based payment. Wang Ming more radically stands on the "result-based revenue share" side. However, the consensus between the two is: the era of purely selling tools is rapidly passing. The difference is only "what to transition to." What Big Tech Can't Do, What Startups Can't Do Wang Ming has worked at big tech and has also started businesses outside several times, so he has a relatively three-dimensional perspective on the topic of "big tech vs. startups." His judgment is: underlying models are not suitable for startups to build. "This is something with extremely high capital density—it's a game between the world's top few companies." But there'sa large number of space at the application layer. Big tech's main business gives it strong cash flow to invest in pre-training, but it also ties it down. "This AI wave is a major technological transformation for humanity, and it will change various business forms and production relationships. If big tech doesn't want to disrupt itself—frankly, it's also hard for it to disrupt itself—then there will be revolutionaries." He said that from the logic of the "innovator's dilemma," it's not just at the business level—including organization, people's interest structures, and mindsets—that all determine that in an era of super-transformation, it's very hard for big tech to completely crush startups or even defend itself. However, he didn't romanticize the situation of startups either. A startup's survival rule is "to quickly complete payment verification at a certainaspect to survive." Without first-mover advantage, there's no data moat; without a moat, you wait to be harvested by big tech. "Even if big tech takes a fancy to you and comes after you later, if you've mastered enough data and vertical model capability, they'll choose to acquire or invest instead." This is quite realistic. When Will Agents Truly Become Ubiquitous? I asked him when agents will truly become indispensable to every person or enterprise. He gave one keyword: Memory. "If Personal Memory can be solved well, all applications will rapidly penetrate into our lives." His logic is: current Chatbot products (Kimi, ChatGPT, etc.) are already quite usable in less rigorous scenarios—collecting information, sparking inspiration, simple Q&A. But they lack continuous memory of the user. Every conversation feels like starting from scratch. If an agent can remember who you are, your preferences, your work context, and where the task you assigned last time stands—its practicality will undergo a qualitative change. As for which industries will be penetrated first, his judgment is: industries with low knowledge density and high standardization will go first. Customer service, legal documents, sales lead screening, emotional companionship—these already have high penetration rates. But fields requiring human creativity, emotional judgment, and intuition will see slower penetration. "Asking AI to do long-chain analysis and decision-making is still relatively not feasible today," he said. "If the chain is serial, with 80% accuracy at each step, after ten steps it's useless. But if many sub-agents collaborate in parallel, each managing a segment, then aggregating—that's much more reliable." He judges that we'll see more practical scenario rollouts in the second half of this year. Words for Ordinary People and Bosses Near the end, I asked him what advice he has for ordinary people and for business bosses respectively. For individuals, he said: "Don't think AI is going to disrupt you—think about how it can help you. AI will most likely help top experts become even better; it amplifies your strengths. Figure out what your strengths are and actively embrace it." For bosses, he was more direct: "A boss'scognition on AI determines the life or death of the enterprise. If you don't embrace AI, the enterprise will die an ugly death—don't complain about the big environment. AI is a number-one-position project; the boss must free themselves up and spend time personally learning and exploring." Finally I asked him: what is the summit K2 Lab is currently climbing? "Hope to become the Agent OS for a certain group of people in the future AI world, helping them manage complex affairs and letting humanity return to creativity and love." He paused, then added: "You don't necessarily have to think about replacing anyone's job—just go spread love." This sentence, coming from a serial entrepreneur, is a bit of a contrast. But I think he means it seriously. Before the live stream ended, he said to the camera: "Physical and mental health—must have physical and mental health! Go climb a mountain tomorrow!" Good advice. After all, no matter where the summit is, you first have to be alive to walk to the foot of the mountain. Selected Q&A Q: Why is it called K2 Lab, not K2 Labs? Wang Ming (Founder of K2 Lab): Labs should be many Labs, right? Actually this is also a reflection of the AI era—the barrier to product innovation has lowered, and one team might genuinely build multiple products; we're already exploring several. K2 is Mount Qogir, in Chinese called Panfeng. I personally like the outdoors and mountain climbing, and I feel that exploration in the AI era is like mountaineering—you need to gaze at that highest peak from afar. Q: What made you decide to leave Alibaba and start a business? Wang Ming: At Alibaba I met hundreds of startup companies. I found first that the basic model capability was sufficient; second that the ROI math worked out; third that startup companies are so fortunate, able to spend 100% of their energy learning and creating. At a big company, a lot of time is consumed internally fighting entropy increase. The evolutionary speed of this era is too fast, and the decisive battle for the final ecological niche will come earlier—I really couldn't stay any longer. Q: Can startup companies build underlying large models? Wang Ming: Not really suitable for startup companies. This is something with extremely high capital density—it's a game between the world's top few companies. Of course, for very small niche-direction general models, there's currently still a window period that big tech temporarily doesn't care about. But that doesn't mean they won't do it. Q: Why choose TikTok commerce instead of enterprise SaaS? Wang Ming: To B business is linear growth—climbing step by step, knocking on doors one order at a time. In China, people feel that interacting with a tool or agent means treating it as a tool, and it's hard to command a high premium. Competition in the AI era will be short and white-hot; we need network effects and need to be closer to money. Q: With performance-based payment, doesn't the company bear the risk itself? Wang Ming: A startup's survival rule is to quickly complete payment verification at a certainaspect to survive. If you don't optimize results, users acquired through ad spend have very low retention. Look at coding agents like Devin—once the model is good enough to become aessential demand, in a few months revenue exceeds the entire previous year. Q: How exactly does the "AI hires people" you mentioned work? Wang Ming: Overseas users are quite lazy; they find interacting with AI troublesome, and even say "why don't you just hire me?" For high-cost scenarios that AI can't handle, we let users handle them. If they're not willing, letting AI handle it is very costly, and they get less money. It's equivalent to them getting a base salary, being hired by AI. Q: Why choose the US market, not Southeast Asia? Wang Ming: Compute costs are expensive, while Southeast Asian local labor is cheap—the math doesn't work out. The business logic of an AI Agent requires being more efficient than人工 and lower cost to hold. The US market has expensive labor, so AI's comparative advantage is more obvious. Q: OpenClaw feels like its热度 is declining—does it really work? Wang Ming: People feel it's not working, but we feel it is. Open-source programmers and major foundations worldwide are all contributing to it, like early Linux. The reason it broke through is that the user friction coefficient became low enough. In the future, if Personal Memory is solved well, all applications will rapidly penetrate. Q: Do your company's employees really code internal systems themselves? Wang Ming: Yes, the HR system, operations system, and financial BI are all coded by employees in each department themselves. We provide unlimited Tokens for everyone to consume, and especially encourage everyone to test the boundaries of the most powerful models currently available. Monthly Token consumption is roughly equivalent to 30,000 employee salaries. The efficiency of managing Tokens determines gross margin. Q: Any advice for ordinary people? Wang Ming: Technological transformation can make people uncomfortable and anxious, but don't think it's going to disrupt you—think about how it can help you. AI will most likely help top experts become even better; it amplifies your strengths. Figure out what your strengths are, actively embrace it, try it. Physical and mental health—must have physical and mental health! Go climb a mountain tomorrow! --- Original publication: https://uniqueresearch.substack.com/p/src-20260516-01html On-site reading page: https://ffcap.cn/en/research/src-20260516-01html