ZhenFund’s Liu Yuan on 5 Bets on Manus Founder Xiao Hong
Summary
Over 9 years, Liu Yuan backed Xiao Hong’s team 5 times, betting not on any fixed direction but on a founder whose judgment kept getting better. From a RMB1M angel investment in Yanyun Technology, to a product with tens of millions of users and an exit at roughly 10x, then Butterfly Effect, Monica, and Manus, the direction changed repeatedly while the core team stayed intact. Liu describes the initial return as “a small angel went in, a big angel came out”; what truly compounded was his understanding of people.
Xiao Hong’s most valuable signal is his willingness, even in good times, to kill a nearly finished product and redirect scarce team attention toward a higher-order opportunity. He abandoned Jianji, a nearly shippable Chinese version of Benchling, after seeing GPT-3’s translation quality and switched to Monica; later, he held back a browser because it was “not different enough,” then launched Manus roughly 2 months later. Liu’s summary is that these pivots were “neither incompetence nor opportunism,” but “Attention Is All You Need.”
Manus’s breakout moment is not yet a success verdict: it is still debating PMF and could be killed quickly by a single product launch from OpenAI, ByteDance, Anthropic, or Google. Liu sees it as a rare product that won international users and strong resonance among mainstream founders from day one in a new form, while repeatedly warning that “all young startups can die at any time.” Cursor and Manus share the same survival mindset: a company’s default is not continued existence, but earning anew each day “a reason to still be alive tomorrow.”
Liu has moved from betting on sectors and polished résumés to reading people from the bottom up: people are harder to change than direction, and they are an early-stage company’s most important and least replaceable asset. He now cares more about whether a founder has sustained passion for a category, can zoom in and out freely between industry forces and individual component suppliers, and can actually execute. He stopped asking “how big is the market” long ago, because user problems, product formation, and the hardest parts of solving them test founders more rigorously than top-down business logic.
This framework came from reversing old mistakes: elite schools, investment banking and consulting, big-tech executive roles, and shared taste with investors had all been given too much weight. In 2016, Liu personally invested in 26 companies and ZhenFund invested in more than 100; around 2018, he also backed 7 or 8 ByteDance executives, only to later ask, “Was it the person who was exceptional, or ByteDance?” He now favors underdogs whose résumés look ordinary but who have depth, leadership, and practical ability, admitting that “shared hobbies are too unimportant—perhaps the more you share, the worse.”
ZhenFund institutionalized FOMO because the biggest angel-investing mistake is not investing in the wrong company, but never seeing the person who might generate a 100x or 1,000x return. Each week, the team spends roughly 4 hours checking highly starred GitHub projects, highly voted Product Hunt products, papers, new financings, and Demo Days, asking “Who invested in the first round, and why didn’t we see it?” Follow-up is tracked in Lark. Liu believes coverage must come before judgment; only then can “missing it” become a lifelong regret.
Liu’s biggest concern now is not that the bar is too low, but that raising it will eliminate the tolerance for error that early-stage investing requires. Unitree Robotics, Chagee, and Pop Mart all had varying degrees of contact with ZhenFund but never became investments, showing that the firm has not only false positives but also more dangerous false negatives. The rearview mirror makes investors demand founders who are “young yet mature, prominent yet undiscovered,” but the high-tolerance investments that looked like blind bets 10 years ago may be the source of today’s best returns.
Deep dive
1. Manus’s breakout moment must coexist with the possibility of dying at any time
When Liu Yuan next met Xu Xiaoping, he had to explain Manus from scratch—what it was and who founded it. He felt Xu’s curiosity was no longer coming from an investor’s perspective, but more like an elder checking in on a student or younger colleague, giving the presentation the feel of “showing a teacher the work.”
Liu admits he gets “visibly animated” when discussing Manus, but cools the temperature himself after almost every section. Tech history is full of products that “opened a new era” only to be forgotten soon after their moment in the spotlight; many pioneers ultimately became merely “martyrs.”
Even if it went no further than today, Manus would still be enough to make an early-stage investor proud: from day one, it entered a new technological frontier and business form, reached international users directly, and generated strong resonance among mainstream founders and investors. Liu likens it to “finally earning a ranking with the swordsmanship you had learned.”
2. The first RMB1M bet backed a student team on the verge of breaking up
Liu says his first investment in Xiao Hong’s team was in 2016, into Yanyun Technology; Koji recalls ZhenFund’s first investment happening in 2015. Xiao and several classmates from Huazhong University of Science and Technology had made money from campus products. When the cash was nearly gone, they were preparing to accept big-tech offers, then went to Beijing for a hackathon simply hoping to win prize money and extend the runway a little longer.
Liu happened to be a judge. What moved him was not the team’s résumé, but the product already built on stage, the way it was presented, and the team’s speed of execution. The young team needed little capital, so ZhenFund quickly invested RMB1M, opening its first institutional financing.
Less than a year later, the money was again running out. Liu introduced Xiao, Huijie, and Laoda to Koji. The 3 spent an afternoon at Beijing’s Shuangjing Apple Community; Koji liked Yiban and transferred several hundred thousand yuan that same day. His view was that it would be “such a shame” if a team like this stopped making products.
These core members would go from Yiban and Weiban Assistant to Monica and Manus. The company’s immediate financial pressure was relieved, but more importantly, the team received encouragement to keep building when it was close to giving up.
3. Yanyun Technology’s modest exit validated the team’s ability to complete the full cycle
Yanyun Technology never followed the typical A, B, C, D financing curve; its fundraising process was “pretty miserable” throughout. But Yiban and Weiban Assistant each reached tens of millions of users. For a team of undergraduates, the product, user growth, and even revenue performance were all remarkable.
The company was acquired around 2022. Because ZhenFund’s principal was small, even a return of roughly 10x translated into limited absolute proceeds. Liu put it vividly: “A small angel went in; a big angel came back out.” It was “small money in, small money out.”
More important for the next investment, Xiao had taken the original team through entrepreneurship, growth, and exit as a complete cycle. When ZhenFund made its second bet, it was no longer buying only the execution of several students, but a well-coordinated serial-founder team willing to start over together.
4. “Build good tools for humanity” turned the post-exit gap into the next startup
Around 2018 or 2019, Liu asked Xiao for his mission statement in life and work. The answer was “build good tools for humanity.” From then on, whenever Liu saw an interesting tool product, he sent it to him.
During acquisition discussions in 2021, the 2 were already discussing whether to start another company. In 2022, Liu began talking with him more seriously about a Benchling-like direction. He described Benchling as “Feishu for scientists,” with a valuation at the time of roughly $5-6B.
Liu’s premise was that tools for intellectually intensive users such as scientists were generally underdeveloped, while Chinese teams had already demonstrated the ability to build complex software products. When Xiao decided to start again around research SaaS, ZhenFund made its second investment “without hesitation”; by the time the deal went to the investment committee, the team had already produced a polished working demo.
5. One GPT-3-translated email rewrote the product roadmap
ZhenFund introduced Xiao to an early co-founder of Benchling so he could learn from the person’s startup experience. Because the co-founder did not speak Chinese well, Xiao used GPT-3 to translate an email and unexpectedly discovered that the model’s translation was already very strong, but lacked a sufficiently friendly user interface.
Xiao saw an opportunity to build a usable “shell” around GPT-3’s capabilities. On November 10, he realized this was the biggest startup opportunity he had seen in AGI; Monica was started on November 20, and ChatGPT launched on November 30.
This pivot was not caused by trouble in the original project. Jianji, the Chinese version of Benchling, was nearly ready to ship, partnerships with big tech, fundraising, and team morale were all going smoothly. That was precisely why Liu saw killing the product as evidence of Xiao’s ability to act immediately after spotting an opportunity one level higher.
During the Spring Festival, Xiao did not even have time to convene a board meeting before using his own money to acquire the ChatGPT for Google plug-in. The name sounded as if it belonged jointly to ChatGPT and Google, though it had nothing to do with either company. Liu joked that its greatest asset may initially have been the name.
6. Team attention is more valuable than a product that already works
ZhenFund’s instinct at the time was that if the team could build a demo in 1 week and quickly make the product usable, why not launch Jianji first and test it? Xiao made the opposite trade-off: team attention was scarce, and “Attention Is All You Need.” A smaller opportunity could not be allowed to hold back a larger one.
ChatGPT for Google and Monica operated, grew, and monetized in parallel, but initially sent no traffic to each other. Monica later did not directly funnel users to Manus either; Liu even half-jokingly urged Xiao, “Send it some traffic, send it some traffic.”
Liu sees this restraint as long-term thinking: let the 2 products grow organically rather than relying on an existing product to drive distribution.
7. Monica’s real growth led ZhenFund to back the same team for a fourth time
ChatGPT for Google quickly captured the window and grew. Monica rose from roughly 3,000 users to several hundred thousand, possibly close to 1 million. Liu explicitly says he cannot remember the precise user count, but revenue had already reached several million dollars.
Monica’s floating widget began appearing frequently on the demo computers of other AI founders. Some internal user groups at large companies reached several thousand, approaching 10,000. It had become one of the products that quickly captured the AI application wave and gained some recognition, and it also appeared on the A16Z GenAI 100 list.
ZhenFund followed with another investment. By the third investment in Butterfly Effect—and the fourth investment in the team cumulatively—Sequoia, Tencent, and Wang Huiwen had also joined. Liu’s reasoning was straightforward: “Monica kept growing.” Some people thought the product category was becoming crowded, but the team and founder were maturing rapidly.
8. A cold browser financing round exposed the market’s discount on existing growth
Xiao was not satisfied with a plug-in and wanted to move up one layer: a browser for the AI era and a new way to search for information. Liu therefore introduced him to Ji Chao of Pick. Ji had built Mammoth Browser at 18, later worked on the search product Magic, and had experienced both fundraising and a company sale.
Investors, however, focused almost entirely on debating whether a browser could work. At first, ChatGPT for Google was treated as a first-stage rocket chasing a hot trend. Even after Monica had large-scale users and revenue, the market still assigned it almost no independent commercial value.
The financing round ultimately included names such as Tencent and Sequoia, but the valuation was only slightly above the round ZhenFund had proactively participated in earlier, despite users growing from roughly 3,000 to several hundred thousand, possibly close to 1 million. Liu recalls that the founder still felt “pretty depressed” when discussing that financing.
9. A browser investors failed to recognize should not blame the users
The browser was actually close to completion. In Wuhan, Xiao demoed multiple features for Liu, who still asked, “Show me the browser now.” Xiao replied in disbelief that all the demonstrations had just taken place inside their browser; for a moment, “the air froze.”
Liu concluded that the problem was not insufficient functionality, but that the overall product was “not different enough.” Arc at least lets users immediately perceive the difference between its vertical tabs and Chrome’s horizontal tabs. If users do not even realize they are using a new browser, the differentiation has not reached the product’s surface.
Xiao had often corrected Liu by saying, “Users are never wrong.” So the browser’s failure to meet the internal launch bar could not be blamed on users failing to understand it; it could only be blamed on the product being insufficiently new and different. The team ultimately did not launch it.
10. Manus retained the browser’s task while abandoning the browser’s form
Roughly 2 months after the browser was shelved, Manus launched in March. Liu believes Manus serves the function of a browser from the perspective of searching for and collecting information, without needing to conform to the form understood by browser purists.
The change was that the team stopped insisting on the shape of a “classical-era browser.” Manus performs the function of a browser, but is no longer a browser in the traditional sense.
Liu’s summary is that neither major pivot came from running out of options: Jianji was killed after the team saw an opportunity on a higher plane, while the browser was not launched because it failed to meet the team’s internal bar. These changes were “neither incompetence nor opportunism,” but the CEO’s most important skill: making trade-offs.
11. ByteDance’s 2024 acquisition offer was a choice about the ceiling of a life
At the beginning of 2024, Monica received an acquisition offer from ByteDance, and Zhang Yiming also met Xiao. Xiao was deeply conflicted about whether to sell. He first had 2 phone calls with Koji, after which the 2 spent almost an entire afternoon talking at a Hong Kong-style tea restaurant in Terminal 2 of Shanghai Hongqiao Airport.
Koji strongly urged him not to sell. The logic was “simple and blunt”: selling might become the founder’s personal ceiling, while keeping Monica and the company left the future potentially unlimited. Monica’s downside was already high enough that even without a sale, it could provide strong financial returns and a solid professional foundation.
Another consideration was that not every founder is suited to becoming a professional manager at a big company after an acquisition. Koji believed Xiao was better suited to being the number one decision-maker and could fully realize his potential only while building a company. For people like that, being blocked in entrepreneurship is painful, but “the emptiness of having nothing to do” may be worse.
In December 2024, around Xiao’s 10th anniversary as an entrepreneur, Shizilukou recorded an episode that has not yet been released. He had already clearly said that he saw the Manus opportunity and was preparing to go all in.
12. Liu’s confidence in the team grew, but he refuses to write Manus as a success story
Asked whether he had ever privately wondered, “What are you doing?”, Liu answered, “Absolutely not.” During the more than 2 years before Manus launched, he recommended Monica to everyone he met and even suggested that ZhenFund interns and graduates spend time looking at the company first.
His ability to persuade Ji Chao to join is especially revealing. Ji had appeared on the cover of Forbes at 18, served as a number one executive, raised money from China International Capital Corporation and Sequoia, and sold a company. Liu had been waiting for him to start again, yet actively pushed him to join another founder’s team.
But Liu immediately put a boundary around that confidence: Manus is “far from successful,” and the team is still debating PMF today. At most, it can be said to have temporarily sparked a wave and begun defining one form of agent interaction. That is “a little bit of credit,” not proof of the final outcome.
13. Chinese VC is very young, but Liu’s first job exposed him to decades of history
From 2011 to 2014, Liu worked at Greenspring Associates, a VC fund of funds. It invested in VC funds as an LP and also selected companies from those funds’ portfolios to participate in or lead later B rounds.
He was the company’s only Asian employee, so he proactively built relationships with Chinese VCs and systematically studied the histories of global funds. In 2017, he also wrote “The Suicide of a Top VC,” discussing Crosspoint, a firm now rarely mentioned.
Many US funds had already reached their 19th or 22nd funds, with histories stretching back to the 1970s and 1980s. By contrast, many Chinese investors still active today on the Midas List and across the industry were part of the first generation of domestic venture capital pioneers. Liu therefore concluded that Chinese VC, despite multiple cycles of rise and fall, remains in “a very young childhood.”
14. The 2016 boom led investors to mistake the tide for discovery skill
When Liu returned to China and joined ZhenFund in 2014, most mobile-internet angel opportunities had passed, but the value of companies such as ByteDance, Pinduoduo, and Bilibili was beginning to emerge in their A and B rounds. Then came the “mass entrepreneurship and innovation” wave, O2O, bike sharing, shared power banks, shared KTVs, and private KTV rooms.
In 2016, Liu personally invested in 26 companies, while ZhenFund invested in more than 100. Companies often raised their next round within 1 or 2 months. Today, a Chinese fund investing in more than 20 companies a year would be considered highly active; at the time, it seemed normal for the industry.
Liu later realized that by the time a sector became a “hot trend,” it was often already too late for angel investors. A wave becomes visible because companies have already succeeded first, after which imitators rush in. The real target should not be the wave riders, but “the people who make the waves.”
He did not chase every hottest deal at the time, but he still suffered from FOMO. More dangerously, many companies that later failed were incorrectly labeled positive examples while the trend was still hot. The supposed patterns investors extracted from them could therefore have been inverted at the source.
15. Business logic did not disappear, but top-down deduction was downgraded
Around 2016, Liu invested heavily in public accounts and content entrepreneurs, believing that “tools turn into communities, and communities turn into e-commerce,” and that large accounts with female, maternal-and-child, or vertical audiences could naturally grow into brands. Xinshixiang was one of the better outcomes, but much of the reasoning did not work smoothly.
The imagination of that era was that seeing a consumer product meant seeing the next Coca-Cola; seeing content meant seeing the next Netflix; seeing a social product for young people meant seeing the starting point of Facebook. Rising optimism and double-digit GDP growth amplified the mood.
Liu did not reject logic as a result. He emphasizes that he is a committed materialist and believes in basic frameworks such as “A is not equal to not-A.” What he no longer believes in is an oversimplified version of “business logic” that tries to fit a chaotic world with a handful of variables and may be about as reliable as “forecasting the weather 2 days from now.”
16. “Investing in people” is the logical conclusion of the early-stage asset structure
Liu’s core assumption is: “People are the hardest thing to change; people are harder to change than direction.” He believes Xiao’s personality, learning ability, and intelligence were harder to change than Butterfly Effect’s direction and product strength.
Startups have few mature assets at scale, while direction can change at any time. People are therefore both the most important asset and the hardest asset to change. Given that structure, “investing in people can only be important; it has to be important.”
He still believes people change gradually rather than suddenly: the person who becomes exceptional in the future should have left some continuous signals in the past. What changed was not this assumption, but which past traits could carry forward into entrepreneurial ability.
Early ZhenFund often treated elite schools, overseas degrees, investment banking, consulting, and big-tech experience as proof that someone was an outstanding candidate. Liu later reflected that a person’s past choices reveal future choices, and joining the safest, most polished institutions such as Goldman Sachs or McKinsey may not be the choice of an entrepreneur like Zhang Yiming, Wang Xing, Bill Gates, or Zuckerberg.
17. Liu’s framework for reading people shifted 4 times—from overseas returnees to content, big-tech executives, and young founders
When he first joined ZhenFund, the team itself was learning angel investing. Xu Xiaoping, Anna, Fang Aizhi, and Yusen were not traditional financial investors. Drawing on the talent network from the New Oriental era, the most natural targets were people from elite schools, fully funded overseas programs, overseas-returnee backgrounds, investment banking, and consulting.
Most of the first batch of polished-résumé companies failed. One company with an exceptionally ordinary background instead kept running: Unitree Robotics. Its founder relied on persistence, conviction, attention to customers, and the ability to identify and solve problems, eventually building a company worth several hundred million dollars. People with more escape routes gave up; he had no other life to switch to.
In 2016, Liu tried to make content his own “superpower,” meeting intensively with public-account operators, major Weibo accounts, Zhihu influencers, and short-video creators. At the same time, he invested in autonomous-driving companies such as Momenta, FinTech, and Perfect Diary, which he backed in 2016, operating in a broad market where it seemed almost anything could work.
Around 2018, he systematically mapped organizations including ByteDance, Kuaishou, Bilibili, Mobike, and Ele.me, investing in 7 or 8 ByteDance executives alone. By 2022, he began shifting more time toward young founders and serial entrepreneurs because “it is hard to separate whether the company was great or the executive was great.”
18. The advantage of an underdog lies in abilities invisible on a résumé
Within ZhenFund, Liu is now often viewed as an investor skilled at finding underdogs, but he stresses that underdog does not mean “grassroots.” It means someone whose excellence has not yet been fully priced in by paper consensus: intellectual depth, ability to execute, leadership, and information processing may all be hidden beneath an ordinary résumé.
Liu observes that Zhang Junjie of Chagee, Wang Ning of Pop Mart, and Wang Xingxing of Unitree Robotics did not fit the polished profile traditionally favored by VCs. One case can be called an outlier; when multiple market focal points repeatedly show similar traits, a pattern may already be forming.
Star serial entrepreneurs usually have a higher probability of success, but they also command higher valuations, face heavier competition, and often do not need the money themselves. Their companies carry too much expectation from birth, making the growth required in every subsequent financing round more demanding. A company “born with a silver spoon” may not end up well, though Lei Jun and others are important counterexamples.
19. Long-term passion and the ability to zoom in and out replaced “how big is the market”
Liu tests passion for a category not by listening to motivation stories, but by looking at time invested: “Companionship is the longest-lasting declaration of love.” Someone who has studied and practiced something over a long period, can talk about it without stopping, and knows every detail is showing evidence of genuine passion.
A good founder must also zoom in and zoom out at the same time: explain industry change, competitive structure, and listed-company gross margins from above, then clarify every component, supplier, and specific user problem from below, moving freely between the macro opportunity and micro execution.
His questions are now entirely bottom-up: How did you come up with this problem? What did you observe? Why choose this product form? What other options did you consider? What was the hardest part? How did you solve it? The questions target ideation and the execution process, not a polished business story.
“How big is the market?” has not been a core question for Liu in a long time. Entrepreneurship starts with understanding users, finding a problem, building a solution, and iterating continuously. A top-down view of market share and competitive advantage cannot replace these facts grown from the ground up.
20. Yusen, serial entrepreneurs, and failure cases jointly opened the bottom-up switch
Yusen had founded a company and led product, so the questions he asked at ZhenFund naturally leaned toward execution and had a major influence on Liu. At the same time, some founders with excellent backgrounds gave up quickly, forcing Liu back to first principles: it was not enough to ask for an attractive entrepreneurial motivation; he had to see how the product was actually formed.
When Liu was meeting young serial entrepreneurs intensively early on, the focus was their previous company. Founders recounting real experience do not start with market size and market share; they talk about concrete problems, choices, and difficulties. A large volume of such narratives gradually shaped his thinking.
The method is not novel. It is practically common sense that appears repeatedly in classic Silicon Valley investment stories. Liu jokes that after more than 10 years, he relearned through “knowing this matter requires personal practice” what had already been written in books decades ago.
Liu also recalled hearing early stories about Zuckerberg being eccentric and unconventional. In industry practice, however, investors had historically favored people who were flawless in business settings and polished in expression, while founders should care most about customers and users—not about making investors comfortable.
21. Sharing the investor’s taste was Liu’s most concealed misjudgment
When investing in content, Liu cared deeply about taste. If an entrepreneur read the same books and liked the same writers and directors, he felt an intuitive connection and unconsciously gave the person extra credit.
He later concluded that this was not a positive indicator. Founders have very little time, and their attention is consumed by building the company; they should not be judged through shared restaurants, wine, books, or other lifestyle preferences with investors. Shared hobbies are too unimportant—perhaps the more shared, the worse.
The deeper calibration was that Liu himself was not a successful entrepreneur, so he could not use himself as the benchmark. He should learn from successful founders he had observed rather than mistaking “like me” for “has potential.” In retrospect, he calls this a “particularly stupid” mistake.
22. After the professional halo of VC faded, what remained was experience and knowing what you do not know
When Liu moved from LP to GP, he viewed industry peers as “godlike”: all-stars who could charge management fees from fund-of-funds investors, while he felt like a rookie entering the field for the first time. After the industry’s romanticization and subsequent disenchantment, he gradually saw that many people had simply been pushed into those positions by the era, just as he had.
When Perfect Diary went public, Liu described himself as “the person sitting in the trunk,” not the co-pilot described by entrepreneurs. In 2013, at 24, he could recite the backgrounds and investments of the GPs at the world’s top 100 VCs because the job had a low barrier to entry but very few practitioners, making information asymmetry unusually valuable.
Liu believes the best VCs have usually participated directly in building great companies. At the next level, they should at least have observed company growth over a long period and at close range, creating a credible playbook. Ronghui adds that a lack of experience is not shameful, provided you “know what you don’t know” and have a method for finding answers.
Ronghui points out that Chinese entrepreneurs can now compete head-to-head with their American peers—6 of the top 10 apps in the US App Store are even Chinese companies. But Liu believes that Chinese funds at the same level still lag US funds by “a very large margin,” both in fund management and individual investing ability.
23. ZhenFund turned FOMO into a weekly coverage system
Liu defends FOMO: in an angel round, a wrong investment can at most lose a limited amount of principal, while missing a deal can mean forfeiting a 100x or 1,000x return and living with the regret forever. For a fund centered on investing in people, the necessary condition before making the right judgment is not intelligence, but seeing the founder in the first place.
Each week, ZhenFund spends roughly 4 hours jointly reviewing the most-starred GitHub projects, most-liked Product Hunt products, new papers, and companies that recently announced financing. The team then asks who invested in the first round, whether ZhenFund had seen the company, and why it failed to find it.
Follow-up items are entered into Lark forms, and the following week the team checks whether contact was made and whether a conversation happened. Shizilukou and Qiji Demo Days also list every project one by one and verify who had or had not spoken with each team, turning every omission into a channel gap to close.
The system covers both inbound and outbound: excellent founders should be more willing to approach ZhenFund, while the team should proactively find people through products, code, papers, and events. Liu does not find the process dull because every 2 years, the people, topics, taste, and beliefs in front of him have changed.
24. Positive feedback can reverse at any time; so-called path dependence may just be an untested style
Liu warns that the positive feedback Manus is generating today may be temporary. If OpenAI launched an innovative product 6 months from now and made Manus disappear, the earlier conclusion that “giants struggle to beat startups through innovation” would be completely overturned by the facts.
Huang Zheng was once asked what Pinduoduo’s end state would be. His answer was only: “The end state is definitely death.” Liu says Cursor’s internal mindset is similar: every morning, the company’s default is that it has no reason to exist and will die tomorrow; the team must earn its right to survive again.
Manus faces the possibility that OpenAI, ByteDance, Anthropic, or Google could produce a competitor the next day. Liu says this is not an abstract risk mentality, but something that could “literally happen tomorrow,” so a breakout moment must never remove the company’s survival default.
Path dependence has 2 sides: if the same method wins on 2 investments in a row, it is called consistency of style; if the second fails, it is called path dependence. Liu admits that after Monica, he became especially partial to Huazhong University of Science and Technology teams, but believes repeatedly asking “How could this be wrong?” can reduce blind following.
25. Technology cycles change, but several foundational founder behaviors do not
Liu borrows Bezos’s thinking: instead of chasing directions that constantly change, look for invariants such as “faster, better, cheaper.” Early-stage investing also needs a set of evaluation standards that can survive consumer, AI, healthcare, and other technology cycles, though it is still unclear whether this “Holy Grail” has been found.
Ronghui further summarizes the invariants he sees: a willingness to take risks, abandon the status quo, and leave the comfort zone; sensitivity to opportunity, seeing it earlier than others and perceiving its shape and scale more clearly.
Ronghui also emphasizes practical ability: a willingness to get hands dirty, move quickly from idea to product, and enter iteration. After going from zero to one, founders must begin thinking about institutionalization and repeatability. These abilities can hold from navigation, railways, and manufacturing through PCs, mobile, and AI.
Liu’s sleep has not worsened because of Manus. His mindset is “hope for the best, prepare for the worst.” Geopolitics, bans, and whether a big company already has a 7-person team are among an almost infinite number of problems that have not happened. He chooses to “tell everyone when a problem appears” and first deal with what can be changed today.
26. False negatives are more dangerous than bad investments; ZhenFund needs to restore its tolerance for error
ZhenFund used to take pride in missing good companies mainly because it had never met them, rather than because it met them and failed to understand them. Unitree Robotics, Chagee, and Pop Mart broke that comforting narrative. Someone at ZhenFund had once spoken with Unitree; Chagee had been recommended by a financial adviser; and Xu Xiaoping had met Pop Mart.
These points of contact show that the firm has not only false positives—believing a founder was strong and discovering after investing that the person was not—but also false negatives, where the person appeared but the firm failed to recognize the signal. The former costs at most the principal; the latter may mean missing a company that defines an era.
Liu worries most that hindsight bias projects today’s mature founder image backward onto the early years, leading investors to demand candidates who are “young yet mature, prominent yet undiscovered.” These conditions contradict one another; the more perfect the standard, the weaker early discovery may become.
ZhenFund’s tolerance for error was high 10 years ago, and peers may have thought it was “investing blindly.” Yet many of today’s best companies came from that period. Liu now keeps asking: Which things should we have done but did not? Which things are we doing without needing to? And “what mistake might we be making right now?”