YouMind Founder 玉伯: Waiting for the Tree to Bloom

Summary

• 玉伯 draws a hard line through AI applications: AI can do only 2 things—improve efficiency or provide emotional companionship. YouMind “is fundamentally about companionship, even if it looks like a productivity tool”: creating and producing something alongside AI is itself companionship. “Whether it ultimately gets published is no longer important; what matters is the process.” He also hopes efficiency-focused Agents become good enough to do everything, saving users time that they can then spend using his product to create together.
• His sharpest take on the current Agent boom is that first-mover advantage may not hold, while retention is the real secret metric. “If they dare publish a second data point—retention—then I’ll believe them”; investors “may not even be able to get that data.” After testing Manus on more than 20 questions he was curious about, he “didn’t know what to use it for.” He also failed to use up the Devin he bought for $500. His verdict: “loud thunder, little rain”—Agent output is fundamentally “just input for you; you can’t use it directly.”
• The fundraising record: he spoke with 80-90 firms and received 8-9 TS, yet the counterintuitive pattern was that the investors trying hardest to understand what he actually did mostly did not invest. BP 1.0 described “paper and pen in the AI era,” moving from paper-and-pen 2.0 toward 3.0. He is “still defining” what 3.0 looks like, so it is no surprise that 80-90 firms did not understand it. Those who ultimately backed him were primarily betting on the person: “玉伯, you’re the first person who has made AI clear to me.” He breaks the startup decision into 3 questions—are you willing, can you do it, and is it worth it—and took an investor’s advice: “Use a pie to draw a pie, not a dream to draw a pie.”
• His product philosophy stacks up a series of non-consensus views: build a project library rather than a knowledge base, treat the ability to forget as a major advantage, and even encourage quick marriages at YouMind. In response to the host’s sunk-cost objection, he rejected the industry consensus that an Agent should “understand you better”: “A conversation with a friend who truly understands you is not necessarily enjoyable”; “talking every day with the worm in your gut has no value.” He also predicted that in another 5 to 10 years—by 2030 or even 2035—Notion might be in serious trouble.
• His view of large tech companies is that those with more than 10,000 employees are 90% alike, with perhaps 10% of the difference coming from the founder. 马云 is a “knight, a liberal-arts student, guided by intuition,” while 张一鸣 embodies engineering-style tool rationality. “I would definitely choose 马云 as a friend, but I might choose 张一鸣 as a business partner.” Senior executives are “very pitiful,” potentially trapped inside information filters. “Every management role eventually rises to the level of its own incompetence; only an incompetent executive is competent.” The root cause of his departure was not being “the final real decision-maker,” while large companies building AI often prioritize their own models.
• His view on pacing is that scaling too early “really is a trap,” and that building a startup is like growing a tree: fertilizing a fruit tree too early can burn its roots. “Many founders know their watermelon is still green, but they rush to cut it open”; when it is not red inside, they switch directions and “start growing potatoes.” He only wants several hundred paid co-creation users—RMB200/year in mainland China and $20/month overseas. Average usage should cover costs. He hopes DeepSeek releases R1 soon and cuts token costs to one-tenth. His timeline is to find PMF next year and complete the first wave of growth from 0.1 to 1.
• His radical honesty includes admitting that, if startup-building were climbing Mount Everest, he has not even reached Camp 1, while he rewrote 90% of an AI-generated podcast draft. With investors, however, he has learned to “lie in quotation marks”: “When I know I’m lying to someone, I’m also being honest.” Many people fail to get things done because they mistake knowledge for action; without lived experience, what knowledge do they really have? His endgame ambition is explicit: “Success must happen on my watch.” “If we’re not going to build something big, should we let Douyin build something big? I want Douyin to stay small and me to get big.”

Deep dive

1. Fifteen Years as a “Middle Manager”: Internal-Network Spokesperson and “Trust Makes Things Simple”

• 玉伯 first corrected the label others put on him: “Calling me an Alibaba executive is a stretch. I was more of a middle manager.” The industry knows him from open-source projects on GitHub in the 2000s and from Yuque in product circles. “Beyond that, I was just a nobody.”
• A typical example of his role as an internal-network mouthpiece came during Alibaba’s years of “war-achievement culture”: “Someone would do something tiny and then send an email praising it to the skies.” He could not stand it and pointed out that the practice violated the company’s values.
• His favorite value was “trust makes things simple.” Mutual trust allows processes to be simplified. “Credit cards are a great invention. What makes them possible is trust—you can use them before you pay.”

2. The Moment He Left: Resigning on His Joining Anniversary and Sending Himself a Message to “Leave Now”

• He stayed at Alibaba for 15 years not because he was happy, but because he still had “hope”: “As long as you still have hope, you stay somewhere. Once everything starts to feel repetitive, you leave.” After the pandemic in 2022, he felt he was “repeating myself.” By 2023, “not leaving was simply no longer an option.”
• The anniversary carried a sense of ceremony: he joined Taobao on April 3, 2008, and decided to leave on April 3, 2023. Many people sent him “Happy 15th anniversary” messages. “But I wasn’t happy. That night I sent myself a DingTalk message: Leave now.”
• The moment he actually walked out of the campus produced “no particular feeling.” Instead, it felt natural—“almost like, finally, I don’t have to come back tomorrow.”

3. An Outsider’s Perspective: His Former Self Is Someone Else’s TV Drama

• Returning to the hill where his primary school had been demolished, he felt as if he could see “another very small version of myself running all over the hillside” long ago. But it was like watching someone else’s story. He had the same feeling passing the Ant and ByteDance campuses: “Someone used to be inside there. It felt like watching someone else’s movie. Those were just a few episodes, with no particularly close connection to who I am now.”
• That detachment became even stronger after he started a company. Before recording the podcast, he spoke with an investor for 2 hours and “also stepped outside myself to observe myself.” His view of time: “Whatever is past is past; it has already entered history.” He cares more about the connection between his future self and his present self. “As for how the present me is connected to the past me, I don’t really care.”

4. The Cost of Honesty and a Lesson from 阮一峰: Don’t Respond

• Speaking the truth on Alibaba’s internal network inevitably attracted attacks. His way of defusing them was again to adopt an outsider’s perspective: “What I wanted to express was in the first post. Once I posted it, other people’s comments had nothing to do with me—they were attacking my past self, whom they can no longer find. The present me might even attack the past me.”
• 阮一峰, known as 兔哥, who had been on the Yuque team, took this even further: “阮一峰 really knows how not to look. Sometimes I still can’t resist checking; he simply doesn’t. Floods can rise as high as they want—it’s no longer his concern. He posted it, and that’s that.”

5. Why Jike Is Better Than Weibo: Deleting a Post Means Judging Your Former Self

• 玉伯 only started using Jike last June, but found its comments “far better than Weibo in its heyday.” On Weibo, reposting was mainly a way to gain followers—“busy noise without substance.” On Jike, “perhaps a third of 100 comments contain something real and can give you insight.”
• He noticed that some people post dozens of times a day but make only a few posts public, using “visible only to me” as a personal journal. He deletes posts too: “Deleting something is my way of criticizing my past self—that post could not withstand my scrutiny today. I might even see the vanity in it.” The host suggested reposting instead, to preserve changes in thinking. He said he would consider it.

6. Be Wary of Every Success Story; Ask Only What Made Sense at the Time

• Discussing other founders’ stories, he warned: “People sometimes look for things that validate their own experience. That requires caution—it may well be bias.” He is skeptical of all success formulas, focusing more on historical failures and the traps people fell into in their own time and circumstances.
• This epistemology may have formed when he was building Yuque: “Technology is a field where you can measure right and wrong. After moving into products, I realized many things have no right or wrong answer. You should ask whether something was reasonable in the context of that time.” Yuque starting To B and then moving toward To C was not a question of correctness, but a choice: “Different roads can lead to different destinations. A may have a beautiful future, and B may also have a beautiful future. Who is to say which is right or wrong?”
• He also offered a non-consensus view: Notion’s success in building revenue through To B does not mean it made the right choice. “Perhaps in another 5 to 10 years—by 2030 or even 2035—Notion will be in serious trouble.”

7. From Graphic Designer to 3 Dreams: Technology, Products, and Freedom

• His career was more winding than his titles suggest. He joined Taobao’s UED team as a graphic designer in 2008, then was “assigned to write code” as a front-end engineer. In 2010, he joined an internal startup competition at Taobao; it “died after roughly half a year.” He then moved to Taobao’s technology division and Java team as a full-stack engineer. He joined Alipay in 2012 and only began putting 40%-50% of his time into Yuque in 2016-2017.
• In 2014, he wrote down 3 dreams: a technology dream—open source, which he has partially achieved; a product dream—to build products he likes and users like, with Yuque “one-third of the way there”; and a dream of freedom—“to do whatever I want without crossing the line.” After sharing them internally, he realized that “one person’s dreams can influence others, helping them find kindred spirits and people like themselves. People really do need dreams.”

8. The Definition of Freedom: Owning Your Attention

• He divides freedom into 2 types: free to do and free from. “Having both at the same time is freedom—but that is still basically an empty statement.” His more practical answer today is control over attention: “People’s attention is not scarce, but it is constantly being interrupted. That is not freedom. Right now we are both immersed in recording this podcast—that is freedom. If we keep checking our phones while recording, that is not freedom.”

9. The Alibaba Docs Merger Battle: First Pushing for a Merger, Then Pushing Against One

• At one point, senior management pushed to merge Yuque with DingTalk Docs. There was even an “Alibaba Docs” proposal: an independent product modeled on Tencent Docs. 玉伯 initially pushed for it. “Looking back, that was mostly my vanity. If we had really done it, we would have died badly.” In the end, he pushed for the products not to merge, and found the final arrangement “quite comfortable.”
• His premortem was that documents and knowledge bases are 2 different categories. Even as Google Docs and Notion grow quickly overseas, many companies still use Confluence for their knowledge bases. A merger would “confuse the needs of documents with the needs of knowledge bases and weaken the value of the knowledge base.”

10. 吴钊’s Mental Model Was Better: Documents Always Follow IM

• He rarely conceded that someone else had the better judgment, but did so with 吴钊: “Within the Chinese system, Alibaba Docs could not exist because Alibaba is not an IM platform. DingTalk Docs can exist because DingTalk is an IM platform.” In the workplace, “a document is fundamentally a communication draft. It is the same thing as a message you send in an IM group.” The first product manager of Feishu Docs also influenced him: “IM messages are fragmented documents; documents are structured messages.”
• Tencent Docs is “Alibaba Docs in a parallel universe,” proving how difficult independent development is: “There is no built-in scenario, so users have to open it deliberately. In theory, it should be WeChat Docs or WeCom Docs, but it never became that. It is awkward.” Tencent Docs may end up pursuing scale and traffic that it should never have pursued.
• But 吴钊 “misunderstood knowledge bases and Yuque, and never really got started with either”—he treated Yuque as another document system. The opportunity is now opening up: “With AI and large models, knowledge bases can absolutely be rebuilt from scratch.”

11. The Root of His Departure: Not the Final Decision-Maker, and Not Free to Choose Models

• His dream of freedom broke on decision rights: “Whether at Alibaba or working on Feishu product features, I was never the final real decision-maker. I thought we should go to A; the president of Feishu thought we should go to B. After a long discussion, I didn’t agree internally but still said fine. That kind of exhaustion is immense.” He calls this part of the large-company disease: “Everything has to be logically correct and data-driven, but those are methodologies. They are not right or wrong.”
• Model choice was another constraint: “A large company building a product will inevitably prioritize its own models. Geopolitics obviously exists, but spiritually, you should still be purer about it—DeepSeek can be used, and overseas products can use OpenAI. Why not? I want to work on global products, but at Ant I could mostly work on domestic business.”

12. What He Learned at ByteDance: 张一鸣’s Tool Rationality

• When he left Ant in April 2023, he already wanted to start a company. “ChatGPT had just come out, and I didn’t really understand it. I went to ByteDance to learn.” His biggest takeaway was the influence, from top to bottom, of 张一鸣’s highly engineering-oriented tool rationality: “If you want to make money, you have to rely on this, at least in the short to medium term. In the long term, perhaps not. There is no right or wrong; it is simply a choice.”
• He was uncomfortable with Douyin: “I really don’t like Douyin. Internally, people would say Douyin helps you open a window onto the world, packaged in a very polished way. I found it a little awkward.” At all-hands meetings, executives would talk about mission and ideals. “It made my whole body uncomfortable because I had no personal experience of it.”

13. The Case for Pitying Executives: Only an Incompetent Executive Is Competent

• Douyin’s executives “may have locked themselves inside an information filter. Much of what they hear is information processed layer by layer by employees. People choose to believe what sounds good—that is human nature, and it is frightening.” He cited Drucker to make sense of it: “Every management role eventually rises to the level of its own incompetence. Only an incompetent executive is a competent executive. Once you become competent, that is all there is to it.”
• The host challenged him: perhaps he was romanticizing executives, and they may hear negative feedback but simply not talk about it. 玉伯 partly accepted the point, but added an observation: “All the good executives I knew left.” Executives cannot tolerate the position forever; they still want to go build something themselves.
• He is glad he was a middle manager who could move both up and down: “If I had really become an executive, I might have ended up trapped there. That would have been my whole life.”

14. Large Companies Are 90% the Same; The Founder Accounts for the 10% Difference

• He reached 2 conclusions at once. At companies with more than 10,000 employees, form, style, and ways of thinking are determined by scale. “Going from large company A to large company B, 90% is the same. The real gap may be only 10%.” Within that 10%, the founder matters enormously. 张一鸣 is rational, data-driven, and detail-oriented; 马云 is a “knight, a liberal-arts student, guided by intuition, and not someone who looks at data much.”
• His trade-off was clear: “I would definitely choose a friend like 马云, but if I were choosing a business partner, I might choose 张一鸣.”
• Why did he leave ByteDance? “If ByteDance were still in the 锦秋家园 era, I would definitely have stayed. Now it looks a lot like Alibaba, and you cannot reach that 10% difference.” By the time he joined ByteDance, it was already the 梁汝波 era and 张一鸣 was gone. “We were all small fry, all just labor.”

15. One Year at ByteDance Equals 3 Years in the Outside World: Transfers Take a Week, OKRs Cover Only 6 Months

• “One year at ByteDance, 3 years in the outside world” is neither praise nor criticism. “It is a feeling.” Transfers happened instantly: “You might be on team A this week and team B next week.” Alibaba was gentler, communicating carefully and respecting people’s views. By the time you actually moved, a quarter might already have passed.
• There was friction over OKRs. He wanted to make 1- and 2-year plans, but was told to “first figure out what you want to achieve in 6 months.” “I knew they meant well, but it was still annoying at the time.” Fortunately, his direct manager when he first joined Feishu “played in a rock band.” “A rocker—who was he afraid of? He shielded me from a lot of pressure, so I was actually quite happy.”
• His view of Feishu’s people was highly positive: “Whether on the front line or in management, they are genuinely very capable. Their average ability may even be higher than at Alibaba.” ByteDance had no training, following Netflix’s “everyone is an adult” philosophy: learn on your own, set your own goals, and drive yourself to deliver. Once hired, people should be ready to work immediately.

16. The 逍遥子班 Reappraised: Organizational Flexibility and the Collaboration Formula

• He once criticized Alibaba’s executive training programs, including the 逍遥子班 and Lucy班, on the internal network for having “people drawing lines around status and hierarchy.” He wondered whether the participants were direct protégés of 逍遥子. Only after going to ByteDance did he understand their value: they were fundamentally social occasions. “You get familiar with each other, business connections emerge later, and departmental walls come down. The so-called departmental wall exists because people do not understand one another.”
• He found ByteDance’s equivalent meetups “dry and awkward.” 谢欣 once convened employees at levels 4-1 and 4-2 who had joined the company, but 玉伯 felt he had no idea what they were supposed to discuss.
• The host added ByteDance HR’s formula for collaboration efficiency: individual expertise × closeness and trust among colleagues ÷ ego. “We often underestimate the intimacy between individuals.” 玉伯 agreed: “It really matters. Without it, things are very difficult.”

17. The 3 Startup Questions: Are You Willing, Can You Do It, and Is It Worth It?

• Willingness is the force of commitment: Who is the user? What problem are you solving? Why does solving it give you your own share of happiness? Looking back, his calmest and happiest moments came from writing, making podcasts, and receiving payments and individual thank-you notes from Yuque users. “That recognition was the most direct source of strength. It was the sense of meaning I wanted.” So “no matter how difficult it was, I had long been certain that I was willing.”
• Can you do it includes finding money and people. He approached fundraising like matchmaking: “If you don’t like me, I don’t like you either. One investor said they would invest in either a white horse or a dark horse. I thought about it for a long time. I am neither—I am a grey horse. So I need to find people who can accept a grey horse.” His hiring standard is that “people should come to YouMind to build a company, not to build 玉伯’s company.”
• Whether it is worth it came from an investor’s reminder: “Stop talking about doing this for 5 to 10 years. Think more about the next 2 or 3 years. Is there a year-end bonus this year? That is very concrete. Use a pie to draw a pie, not a dream to draw a pie. The pie has to be breakable into pieces that everyone can eat, or the team will not last.” He admitted that he had considered this question but had not taken it seriously enough, approaching it more from idealism.

18. 80-90 Institutions, 8-9 TS: In the End, They Invested in the Person

• The hardest period ran from May to roughly July: “I spoke with 80-90 firms and was close to breaking down.” In the end, he received 8-9 TS, or roughly a 10% hit rate. The counterintuitive pattern was this: “The investors who worked hardest to talk to me and figure out what I actually wanted to do basically didn’t invest. I didn’t explain it clearly, and they couldn’t see it clearly either—so what were they investing in?” The people who backed him were primarily investing in the person, or in the person and the team.
• What moved the angel investor was not the BP: “He said, ‘玉伯, you are the first person who has made AI clear to me.’” 玉伯 used analogies to explain the differences among GPT, Claude, Gemini, AI coding, and Cursor, as well as what AI can do and where its limits lie.
• Would repeated rejection become internalized as self-doubt? “One person questioning you is nothing. Dozens of people questioning you cannot be nothing.” But his outsider’s perspective helped: “Investors, including me, are all reactions to a particular stage of this wave. Their doubts did not eliminate my non-consensus; they deepened my thinking. Beyond the positive and negative sides, people challenging you from 360 degrees are contributing perspectives.”

19. Paper and Pen in the AI Era: From 2.0 to 3.0

• The abstraction in BP 1.0 was this: paper and pen 1.0 are physical paper and pens, which have remained valuable for thousands of years. Version 2.0 is everything with an input method—Word, Notes, Notion, and Yuque all qualify. Version 3.0 is AI writing: “In the AI era, why should we still use an input-method-based model to write an article?”
• What does 3.0 look like? “I am still defining it. That is why it was perfectly reasonable that 80-90 investment firms did not understand it last year.” He has barely changed BP 1.0. “The BP 2.0 I am making now is definitely much clearer.”

20. AI Can Do Only 2 Things: Improve Efficiency or Provide Emotional Companionship

• His overall taxonomy is simple: “AI can do only 2 things. It can improve efficiency or provide emotional companionship. There is nothing else.” On the efficiency side are logical, measurable fields such as law and AI coding, where AI will eventually outperform humans. He is most excited about AI for Science: “I studied physics, and there are still many unsolved questions in physics that I care about.”
• YouMind occupies an unusual position: “It is fundamentally about companionship, even if it looks like a productivity tool.” He broadens the definition beyond talking to DeepSeek late at night when feeling low: “When a person and AI do something together and produce something together, that is companionship. Whether it ultimately gets published is not important; what matters is the process.” This incremental emotional value “comes from the interaction between silicon and carbon.”
• His strategy is to coexist with the efficiency camp: “I really hope this wave of efficiency-focused Agents gets everything done. The better they become, the more time they save users; the more time users have to use my product, and for us to create something together and enjoy the process.”

21. Everyone Can Create: The Photography Analogy

• His 8-character mission is: “Everyone can create, and every act of creation has meaning.” In response to concerns about barriers, he compares it with photography: “Photography has barriers too. Becoming a photographic artist has barriers, but the fact that everyone can take photos only emerged in the last 10-plus years.”
• He acknowledges that his vision shares a source with ByteDance’s: “Inspire creativity, enrich life is a very good vision, and I identify with it.” But when he uses Douyin, “it does not feel like that.” The host agreed that Douyin has given people at the bottom of society “a tool to record and share their lives and be seen. That is a very significant thing.”

22. Quiz-Style Due Diligence and “Lying in Quotation Marks”

• When investors asked how he would drive growth, he answered, “I will personally get in the arena.” Those 4 words may not have appeared among their answer choices, leading them to wonder whether he had no sense of how things worked. “Any question for which they already have an answer in their heads, I answer badly.” His reflection was that the question itself was flawed: “They know I am a first-time founder, yet they ask me how I will do operations and growth. I have to cross the river myself, like the little horse in the fable.”
• The dark humor of his growth was this: “So now I also lie to investors—in quotation marks. I know what answer they want. When I know I am lying to someone, I am also being honest.” The deeper point is that “many people struggle to get things done because they mistake knowledge for action. If you have not lived through it, what knowledge do you really have? It is all fake. What you think I know may simply be something I picked up from reading someone else’s materials.”
• He also gives credit to those who did not invest: “They have their reasons, and I have mine. Everyone has their own rationale.”

23. A Product Is an Environment: Be the Owner, Not the Slave

• His product philosophy is that “a product is an environment.” 张小龙 built WeChat around the principle that mini programs should leave once they have been used; there is design behind all of it. “People ultimately are reactors to their environment.” What troubles him is that “many products are trying to turn ‘people are reactors to their environment’ into ‘people are slaves to their environment.’” In the environment he wants to build, “people are the owners.”
• This led him to an operational definition of freedom. The first step is noticing that you have a choice: “When Douyin serves you a video, you can choose not to keep scrolling. Why do you go home, lie on the sofa, scroll for half an hour, and then blame yourself? Many people do not even notice that they have a choice.” The second step is having a good environment that guides you toward choices that are better for you.
• The feeling he wants to recreate is “a smile of recognition”: as a child in the countryside, after finishing farm work, he would lie on a bench in the courtyard under a sky full of stars, listening to adults talk as the wind passed by. “That is hard to reproduce in the short term. I am building a small environment, and it is affected by the larger social environment. 即刻 has built a small environment where everyone believes in AGI, but the real large environment is 新闻联播. I want to serve people who use AI to write their first Xiaohongshu post or make their first podcast. That is hard—really hard.”

24. Product Status, by His Own Admission: “We Haven’t Even Reached Camp 1”

• Version 0.1 was an “AI folder.” Version 0.2 upgraded it into “an integrated creative environment for content creators,” combining research, note-taking, and AI processing in one place. “Saving users 1 or 2 steps makes things much easier, and the compounding benefit of combining tools and functions starts to appear.” For example, a custom AI assistant can look at a user’s recent drafts and determine which sections of a public WeChat article they should read.
• His own use case was unvarnished. For the third episode of his podcast, he had AI choose a topic and write a 10-minute first draft based on the transcripts from the first 2 episodes. “AI labored away and finished it. I read it and fell silent. It was very average. I reworked it, changing 90%.” When the host asked whether that meant the product was still immature, he answered directly: “It is immature. If this were climbing Mount Everest, I haven’t even reached Camp 1.”

25. Build in Public: Several Hundred Paid Users as Fellow Travelers Filtered Out

• His approach is to build in public: “Even before I have written the code, I put the idea out there. I am not the only one accompanying this product as it grows. That is the open-source spirit in my bones.” The waitlist received more than 10,000 applications, several thousand were approved, and there are several hundred paid users: RMB200/year in mainland China and $20/month overseas. Since last December, he has mainly done simple operations on 即刻 and Twitter.
• His response to the illusion that fans are paying was measured: “I understand it clearly. One wave is moral support; the other is early adopters, the same group of people who use Manus. They are certainly not paying for the product. They are putting down a deposit for curiosity.” Charging is itself a filter for finding seeds of co-creation. He deliberately chose 即刻 for operations to avoid his familiar circle of friends; some users he later met did not know him at all.
• There are 4 feedback channels: one-click feedback inside the product; manual monitoring of public sentiment on 即刻 and Xiaohongshu; overseas Discord groups and domestic WeChat groups—“There can be hundreds of messages in one night. I really want to connect WeChat to AI and have it summarize my unread messages”—and long emails “teaching 玉伯 how to build YouMind.”

26. 3 Non-Consensus Views: Forgetting Is an Advantage, Encourage Quick Marriages, Reject “Understands You Better”

• The host’s real objection was that continuously feeding an Agent a knowledge base creates enormous time and sunk costs: “If you choose wrong, it is like marrying someone and then discovering you made the wrong choice.” 玉伯’s answer went to the root: “That is exactly what YouMind is not very worried about. In fact, YouMind encourages quick marriages. My logic is not a knowledge base; it is a project. This year’s Double 11 has little to do with last year’s Double 11. A lot of so-called knowledge bases are disposable. The ability to forget is a huge advantage for human beings. Around a project deadline, the future matters far more than the past. My current provocative view is that the things in the past really do not matter very much.”
• He directly attacked the consensus that an Agent should “understand you better”: “A conversation with a friend who truly understands you is not necessarily enjoyable. Talking with a stranger is fun and produces growth precisely because your perceptions differ. If you chat every day with the worm in your gut, it is just your worm. What value does that have? Why does it need to understand you?” His reference points are psychologists and coaches: “The point is to stimulate the other person. It is better if they do not understand you, because they can be more objective and know how to draw out your potential.” The AI he is building is “at the YouMind level, not the user level”: the more users it has, the better it understands creators’ pain points and how to resolve them.
• But the direction has not converged, and he does not hide that: “A co-creation partner, a psychologist, or a coach? That is the question I am thinking about. I have not settled on the specific direction from Camp 1 to Camp 2. Come use it first. You may have a good idea in the process—tell me quickly.”

27. Agent Demystified: Retention Is the Secret Metric; He Wants DeepSeek R1 to Cut Costs Again

• His hands-on verdict on competing products was specific and harsh. After Manus “ran through the more than 20 questions I was curious about, I did not know what else to ask. The biggest problem was that I did not know what to use it for.” He asked it to research YouMind’s competitors—Notion, NotebookLM, Flowith, Qwen, and others—and “after reading it, I felt as if I had not read anything.” He bought Devin for $500 early on, “did not even use it up, and it did very little.” His overall verdict: “Agents currently produce loud thunder and little rain. Whether it is a report or a website, it is all input for you and cannot be used directly.” Cursor, which can actually write code and get work done, is better.
• On first-mover advantage, he said: “If they dare to publish the second data point in this wave—retention—then I’ll believe them. I will try anything that costs $500 or $1,000 out of curiosity, but after the trial I certainly will not renew. If people keep using it in the second and third weeks, then you have a real flywheel.” Retention is a secret metric for AI startups, and investment firms may not even be able to obtain it. He is not chasing scale: “I only want those several hundred users.”
• The product connects to SO, Claude, OpenAI, Gemini, and DeepSeek. “The logic is simple: whichever works best.” Heavy users cost more than $20/month, but the average user can cover the cost. He will soon switch to a credit-based system, with users buying additional usage packages after exceeding their allowance. He hopes DeepSeek releases R1 soon and drives token costs down to one-tenth. He wants model providers to compete until AI becomes as fundamental as utilities: “Water is still pretty expensive right now.”
• He is clear-eyed about pricing: “In theory, products should charge based on outcomes. But only a tiny number of very lucky products can do that. If an image-generation product makes you one image, the cost has already been incurred. Even if you dislike all 10 images, you still have to pay. Most AI products today charge by usage.”

28. One Year into the Startup: Less Good at What He Thought He Was Good At, and Growing Fond of Uncertainty

• First realization: the chores he thought he would be bad at, such as finance and legal work, “have all been manageable.” As a result, “I am no longer afraid of many things. In my mind, it has become a world where there is almost nothing I cannot do.”
• Second realization: the veteran product person has been humbled. “A team member born in the 2000s discussed a feature design with me, and his idea was better than mine. We should have listened to him.” Users said they did not want to read a 10,000-character article in full; it should be distilled according to their individual interests into 5 minutes. He connected this with his own experience of reading Dream of the Red Chamber 6 or 7 times: “Real reading does not mean starting at chapter 1 and reaching the final chapter. It truly begins when you have a question.” The host compared this to Lin Daiyu: people assume she had low emotional intelligence, but she had “very high emotional intelligence and strong boundaries. When she wants to make others happy, she makes them happy; when she wants to sting, she stings.” She embodied both free to do and free from.
• Third realization: “I went from being anxious about uncertainty to actually liking uncertainty quite a bit.” He knows that many things are simply a matter of time. When an investor worried about why he had only several hundred paid users, he replied that even excluding the Spring Festival, only a few months had passed. Having several hundred paid users after a few months “still gives me a strong sense of happiness.”

29. Do Not Cut Green Watermelons: “Success Must Happen on My Watch”

• What frightens him most is a mismatch between understanding and action. His experience building Yuque and working on Feishu taught him that “scaling too early really is a trap.” Chasing rankings brought in tens of thousands of visitors, but not even 1,000 users stayed. “You start doubting yourself.” His signature metaphor: “Many founders know their watermelon is still green, but they rush to cut it open. When they see it is not red, they say the direction is wrong and immediately stop growing watermelons and start growing potatoes. They spend the year cutting watermelons. It is a real waste.”
• His gardener’s mindset comes from experience: “My family once had an orchard. I know very well that fertilizing a fruit tree too early can burn its roots.” Bananas sold in cities are picked green and ripened in transit; they are completely different from fruit ripened on the tree in its place of origin. The users he is recruiting “are willing to wait for the tree to grow, wait for it to flower and bear fruit. Only then can they taste how good the fruit is.”
• He closed on ambition and timing. His non-consensus views “will eventually become consensus, but by the time they do, the opportunity may no longer be mine.” So “success must happen on my watch. Without that desire, why start a company? You could simply wait and use it.” His target for next year is for version 0.1 to reach PMF and enter the first wave of growth from 0.1 to 1. His principle is “slow is fast,” and the team has only one cultural word: honesty. “Users can feel the code written by passionate programmers.” Having raised funding to start a company, he is certainly not aiming for something “small but beautiful.” He wants to build something big and strong: “If we’re not going to build something big, should we let Douyin build something big? I want Douyin to stay small and me to get big.”