Kill the Candlestick? This Startup Wants to Make the Stock Market "Speak Human" | BlueRun Ventures Family Headlines

Vakee (Yunqi Lai), founder of RockFlow, an AI-native fintech platform, recently sat down with Huxiu for an interview.

As RockFlow's earliest and largest institutional investor, BlueRun Ventures has backed the company since its 2021 angel round, participated in multiple subsequent rounds, and continued to invest in this round — a steadfast long-term supporter.

In BlueRun's view, what makes RockFlow truly scarce is that it accomplished what almost no one else in the industry dared or was able to do: a ground-up rewrite of the underlying infrastructure, creating a genuinely AI-native financial operating system.

"Build an all-in-one AI-native trading platform." On day one of her startup journey, Vakee wrote down her five-year plan.

In Vakee's vision, the future of investing shouldn't be just cold candlestick charts — it should have the passion and adventure that belongs to young people. The data is already validating this: a J.P. Morgan report shows that Gen Z's investment participation rate at age 25 has reached 37%, far outpacing previous generations — investing is no longer "a game for grown-ups." As digital natives, this generation embraces AI and seeks digital financial companions who resonate with them. RockFlow was built for exactly this purpose.

RockFlow launched in 2022. In September 2025, Bobby, an AI investment trading agent, was introduced as an embedded conversational agent within RockFlow.

Right before our conversation, RockFlow had just wrapped up a "real money, real market" US stock experiment: 10 large language models, each given a $100,000 account, trading 10 of the market's most-watched stocks — NVIDIA, Tesla, Google, and others — making trading decisions every five minutes. All models received real-time data through RockFlow's AI investment trading agent, Bobby, and executed trades accordingly.

"The core idea was to show people how AI trades in real markets, to explore the boundaries of AI applications, and to lower the barrier for ordinary people to participate," Vakee told me about the motivation behind the experiment. The trading-grade data and financial engineering capabilities provided during the competition were the result of four years of accumulated work by Vakee and her team.

But for this market veteran, the competition was only phase one. She describes the platform's ultimate goal as a "strategy marketplace" — ordinary people can tell Bobby their worldview in plain language, and the agent will generate and match the most suitable single or multiple strategies.

"When I say 'buy a little,' and you say 'invest a little,' they definitely mean different things." This is the personalization Vakee wants to emphasize.

Vakee firmly believes in investment freedom — everyone has their own investment philosophy. She even launched a podcast series interviewing 100 investors on the RockFlow platform who've achieved 100x returns. Here, everyone has a different worldview that shapes their investment philosophy. One investor, a post-95 who never studied finance, achieved 100x returns through his own investment logic.

This is the story Vakee has always wanted to tell. This idea of "financial democratization" is fundamentally similar to the "short-video democratization" that Kuaishou advocates.

Vakee, photo provided by RockFlow

After graduating from Imperial College London, Vakee first joined a London-based quantitative fund, then went to Baidu, and later spent some time doing VC investing. Every time I've met Vakee, she's had the same minimalist look — white T-shirt or hoodie, jeans, sneakers, clean and sharp.

Trading on intuition, but behind that intuition lies abstract reasoning from unstructured experiential data.

During her years as a founder, Vakee made two of her most critical investments: after GPT-3.5's release, she continuously added to her NVIDIA position, and at the end of 2023, she went heavy on Robinhood when it was trading at just eight dollars. NVIDIA in one hand, Robinhood in the other — her holdings represent Vakee's vision for the industry's future.

She told me, if you ask where RockFlow is headed, she would definitely say it's more likely to resemble Robinhood than anything else. Founded in 2014, this US trading platform became one of the most active platforms among American millennials in a short time through zero commissions, gamified interfaces, and low barriers to entry. Its market cap has surpassed $88 billion, with its stock price stable around $100 and year-to-date gains approaching 150%.

What Vakee is building, more precisely, is an AI-native Gen Z trading platform — something for which there is currently no reference point.

In our conversation, Vakee showed unwavering conviction about AI-native architecture. In her view, anything that can complete a specific task node can be called an agent, and Bobby is composed of thousands of agents. RockFlow doesn't build foundation models; whether it's DeepSeek, Claude, or GPT, their capabilities vary greatly for different tasks. RockFlow spent two years building a multi-agent framework for the financial domain.

After trying out RockFlow's investment assistant Bobby, you'll find it's a product with very long interaction chains — after you complete one step, the product prompts you to move to the next.

I asked, "If Silicon Valley is saying Scaling Law is ending, is it time to bottom-fish NVIDIA?" Bobby gave a series of suggestions, then you could ask Bobby if it could provide a corresponding AI portfolio recommendation. Bobby then offered an "AI full-house" portfolio, all in plain language, not a single industry jargon term, very easy to understand. It even included a "tip" feature — if Bobby's advice was good, users could tip it.

Vakee's take on this format: if people are willing to pay based on results, that's "pay-for-performance," a more AI-native business model.

On RockFlow, you can also view other people's portfolios. Users can share their decision-making process and conversations with Bobby in the community, and there's even a "copy top performers" feature that enhances user stickiness.

Speaking of ARR (annual recurring revenue), Vakee doesn't find it particularly meaningful. She cares more about revenue per employee and profit margin — generating greater value with fewer people, which is how AI-native organizations can achieve extreme efficiency.

RockFlow is also an AI-native company. Vakee says they've already achieved extremely high human efficiency, and the core reason is that AI makes talented people even better. Theoretically, at RockFlow's current scale, they should have 20 times more people than they do now, but for an AI-native company, that wouldn't make sense.

During our conversation, RockFlow had just completed a new round of financing worth tens of millions of dollars. This round was led by Ant Group, with BlueRun Ventures, Monolith, Forwest Capital, and Evergreen participating.

We also spoke with an investor from BlueRun Ventures, one of the participating investors. The investor said, "What makes RockFlow truly scarce is that it accomplished what almost no one else in the industry dared or was able to do: a ground-up rewrite of the underlying infrastructure, creating a genuinely AI-native financial operating system." The investor added that what will define the quality of AI agents in the future is not how well they answer questions, but how deep they can go in specific scenarios and how many real tasks they can take on. Bobby is proving this point — users can complete the entire process from analysis to order placement with a single sentence. This "scenario closed-loop capability" determines the product's boundaries.

For Bobby's future, Vakee has her own plans. For example, it should be imperceptible — often not requiring deliberate interaction, appearing only when user behavior and needs change. If your Pinduoduo usage suddenly doubles, Bobby might proactively ask whether you want to consider positioning in Pinduoduo stock, which is currently at a low. "Bobby will convey more objective trading patterns, not subjective worlds — the subjective world is each user's own expression."

This is, of course, another approach that matches Vakee's lifestyle and values — trading is a way of life, and since Bobby is an investment assistant, it should be embedded in life. For example, noticing significantly more people around you using flash sales, and immediately asking Bobby to analyze trading strategies for Taobao and Meituan.

When asked about future challenges, Vakee said compliance, because global financial compliance is extremely difficult. As for the company itself, the worst-case scenario would be becoming a very profitable large-model quantitative trading team. Vakee tells her team, "We can always become a 'Renaissance Technologies,' but that's not why we started this company. RockFlow must be a platform company, doing the most challenging things."

Huxiu: Why did you choose to run this AI stock trading experiment right when you announced your funding round? As the organizer, were there any particularly interesting findings?

Vakee: There were many interesting phenomena in the competition. For example, general-purpose models performed very differently in digital assets versus traditional securities markets, which validated their adaptive differences when facing different market rules.

The core reason for launching this competition was to show people how AI trades in real markets, to lower the barrier for ordinary people to participate, and to let everyone generate their desired trading strategies through simple interaction with Bobby.

So for this competition, we didn't intervene in any strategies. We only equally informed all foundation models of the basic US stock trading rules — such as whether leverage is allowed, trading hours, and so on — because these general-purpose models weren't trained for financial trading in the first place. Additionally, we provided the most basic data, such as market data and real-time news on trading targets. After they equally received this most basic information, the models made their own trading decisions.

Every foundation model has its strengths, weaknesses, and areas of expertise. The team will use data generated from each competition to continuously train Bobby, integrating the advantages of different models (Bobby itself already incorporates numerous foundation models), ultimately forming a more mature AI trading solution.

Eventually we'll create a strategy marketplace where anyone can generate a quantitative strategy with a single sentence to Bobby, and simultaneously get matched with the strategy best suited to them on the platform.

Huxiu: When you say "strategies," are you referring to what you'd call a "worldview"?

Vakee: Exactly. Everyone has their own values and perspectives, and when these map onto investing, that becomes your strategy.

Huxiu: But this strategy is AI-generated for you. Does Bobby translate this worldview into a strategy?

Vakee: This is just the beginning. We hope to validate this model through one competition after another. Many people ask me why I don't just do investing or asset management myself. It's not that I can't, but that wouldn't be fun enough for me. I want RockFlow to become the world's largest asset management platform, without us writing a single line of strategy ourselves.

Huxiu: This is actually a very AI-native idea.

Vakee: Right. We haven't explained this much externally, but we're doing the same thing as Kuaishou — lowering the barrier to content creation to the absolute minimum. Kuaishou let everyone shoot videos and be seen; we want to let everyone create strategies, with every strategy deserving to be matched. I don't think everyone needs to do quantitative trading. Even simply saying "I'm just going to all-in Tesla" — that's my strategy.

Huxiu: How did this AI-native idea iterate step by step?

Vakee: I'd been thinking about this before I even started the company. I felt this was what I truly wanted to build. Of course, there's also the critical asset of data. I also thought about how to use technical means to assign value every time data gets used. For example, if one of my ideas requires data provided by Xiaoming, then the profits from that strategy should be shared with him.

Huxiu: Did you encounter skepticism when fundraising in 2021? AI investing wasn't as widely understood back then.

Vakee: It went fine actually — I raised $10 million on day one. Investors may not have fully understood what I wanted to do, and to be honest, my own business reasoning wasn't as clear then as it is now. But we all knew: this is a $100 billion or even trillion-dollar opportunity. Finance itself is a massive industry, and the most valuable thing is financial trading platforms. The new generation of young people has clear new demands, and my two goals of "all-in-one, full-category" and "AI-native" — they understood those. That was enough.

They were mostly betting on me as a person. I'd been in the industry for nearly ten years, whether doing product and R&D at Baidu, or at Baidu's investment department, then moving to VC. People who'd worked with me generally recognized my capabilities and judgment. They believed I could see and go all-in on major opportunities of the era, and were willing to bet on me. I'm deeply grateful for their trust.

Huxiu: Was GPT's launch a critical inflection point?

Vakee: Yes. In my early thinking about an AI-native trading platform, I started with personalized experiences. The emergence of large models most directly lowered engineering implementation costs. Without large models, we'd have had to hire many NLP engineers at very high cost. With large models, most capabilities could be achieved through them — token costs are far below human labor costs, and they're continuing to drop rapidly. And as large models keep iterating, they've created many product capabilities I hadn't originally planned for, such as Bobby's capabilities and agent product forms, bringing more possibilities. Product companies like us are essentially "riding a boat" — foundation models are the seawater, and as they upgrade, we get lifted up, riding the wind and waves together to explore new worlds.

Huxiu: Data is critical. Often, accumulated vertical domain data is a barrier that's hard for AI products to breach, especially for financial products. Compared to large model companies or general-purpose agents, what is Bobby's advantage?

Vakee: There are essentially three core categories of data. The first is trading data itself — financial engineering data, such as real-time quotes, volume and price data, earnings reports, and so on. This data volume is massive. First, it's expensive. Second, obtaining it requires financial qualifications. Third, cleaning it requires sustained long-term investment, relying on financial engineering teams to continuously refine it. We have to process this data very meticulously ourselves, since users need it for everyday trading.

The second category is data related to investing and trading. Social media data, financial news, and data from special scenarios. This data needs to be processed around the core of "serving trading." General-purpose models or general-purpose agents won't specifically process this data — the requirements for processing this trading-grade data are high, the difficulty is great, and the ROI is low. So this is a business choice.

The third and most important category is personalized user data. You have to know a user's asset situation and trading behavior to provide personalized investment and trading services.

Bobby is personalized. If I tell it "buy a little" and you tell it "buy a little," the results will definitely differ. It knows that my "buy a little" means buying 5% or 1% position, while it knows another user's "buy a little" might mean buying 1 share or $100 worth. This all comes from continuously learning user behavior. And only by having access to users' real-time data can recommended strategies be executable. Without understanding user data, you simply can't do valuable, executable communication and recommendations.

Huxiu: That's indeed a critical point. Another key issue is hallucination. Finance is an industry with very low tolerance for error. How do you avoid hallucinations?

Vakee: The pursuit of precision in financial trading knows no bounds. Our solution deeply decouples and reconstructs large models' semantic understanding capabilities from rigorous underlying financial engineering logic.

More importantly, we control all core processes including trading and risk management through strict business rules and risk control systems.

Huxiu: Bobby launched in May this year. How did you do the cold start? Any special marketing moves?

Vakee: Mainly through RockFlow's existing user base. Bobby was initially embedded within RockFlow, first opened up to existing users. They already had trading data, so they could get more precise personalized experiences when using Bobby. These were our first core users. The next phase will gradually expand to new users, finding more "aha moments" — that's our current focus.

Huxiu: Since Bobby's launch, has there been any moment when you saw sudden explosive data growth, or clearly felt the feedback of a data flywheel?

Vakee: Bobby is a product in a new domain, so we're not being too aggressive. Compared to internet products, we need more time — we actually only did public beta in September. But there have been some interesting findings that show the rudimentary form of a flywheel. For example, we've found that more and more users are completing the entire process from spotting trading opportunities, analyzing them, to finally executing trades, all within Bobby's conversational interface, without using RockFlow's other features.

Huxiu: In designing Bobby, what subtractions or additions did you make?

Vakee: The core logic of additions and subtractions is figuring out what users actually need in investment and trading scenarios — defining and positioning the demand is crucial. I consider two dimensions: first, what should the future financial agent look like, what product form should it take; and second, what are users' core needs right now.

Let me start with the future. I believe the financial agent of the future will even be imperceptible. It integrates into your life, silently solving many problems for you, mostly without requiring deliberate interaction. Only when your behavior or needs change will it come to confirm with you. Actually, financial behaviors like investing, insurance, and cash loans are already part of life — the agent should be able to sense your needs. For example, if you're applying for a Schengen visa to Europe, it knows you need accident insurance and will directly match it for you. Your annual consumer critical illness insurance is about to expire, and it will proactively remind you or handle it. It captures your continuous behavioral patterns in life, then combines them with your real situation to provide services.

Huxiu: That's actually a subtraction mindset.

Vakee: Right, the future is definitely about subtraction. Like the best assistant isn't someone you need to instruct daily, but someone who understands you without you saying anything and arranges everything. For example: if your Pinduoduo usage jumps from 20 times to 40 times this month, Bobby might proactively ask you, "Pinduoduo's stock price has been low recently — want to consider positioning?" It matches your investment needs through data from your life. This will take time, maybe three to five years, even ten years. Returning to the present, the current stage is about building user trust, which is also the hardest part. It's not just us — conversational large model products like GPT, DeepSeek, and Moonshot AI are all collectively cultivating user habits and building this trust.

But some areas require addition, such as investment research. Before Bobby, users had to tap through the app one by one to read earnings reports and look up data — it was troublesome to find things, and they couldn't follow up with deeper questions. Now through conversing with Bobby, users can keep asking "What does this data mean? Is the current price reasonable? How are other top investors operating?" and even have Bobby recommend investment portfolios. This is much richer in judgment dimensions than before, but more efficient for users.

So addition versus subtraction is dynamic — it varies by segment and by user needs. The core isn't obsessing over adding or subtracting, but solving users' specific problems right now.

Moreover, the biggest problem in finance is that it doesn't speak human language — the comprehension barrier is too high. I myself don't know what I'm buying when I purchase insurance. I wanted to change my daily transfer limit in a bank app and couldn't find it after searching for half an hour. Finance is clearly a strong, rigid demand for everyone, but precisely because the terminology is obscure and operations are complex, this demand goes unmet. The subtraction we need to do is fundamentally about lowering this comprehension barrier, translating both financial terminology and user expressions into language everyone can understand. For example, many users won't say "I want to modify my daily transfer limit" — they'll just say "I want to transfer more money out each day, how do I do that?" Bobby needs to understand this kind of plain speech. That's the key subtraction.

Huxiu: That's quite interesting. How is the business model planned?

Vakee: We'll move to a subscription model going forward. The habit of subscribing to AI tools has already been cultivated in many countries.

Huxiu: Looking back at the beginning of your startup, is there anything you particularly regret? If you'd known then, would you have done things differently?

Vakee: It's actually fine — broadly speaking, I think it's OK. For the vast majority of things, if I did them again, it would mostly be about optimizing details. For example, if something is at 70 points now, could it be 75 or 80? I think so. But if there's something that's completely failing at 10 points now, and doing it again could get a very high score — that kind of situation is rare.

A lot of those critical decisions were actually the right ones. Let me give you a few examples.

First, the choice of business entry point: At the time, I firmly chose the more difficult path — traditional equity assets in the US stock market, rather than emerging digital assets. Emerging assets are easier to scale quickly, but traditional markets present far greater engineering challenges. Our goal was an "All-in-One" multi-asset platform, so we had to build solid capabilities in compliance and complex trading first. Once you start with the easy stuff, it's hard to get the team to go back and tackle the hard stuff later — both the technology and the team's morale suffer.

Second, the decision to build our own trading system: We made what felt like a "counterintuitive" call — deciding to build our own trading desk from the very early days of the company.

When we started, user demands had shifted. We needed to be all-in-one, and we needed an AI-native experience. These two goals forced us to master the very core of trading infrastructure at the deepest level. Building our own desk became the only viable option, even though it was a massive investment and an extremely difficult deployment challenge for a startup team.

The third major decision was whether to start building large-model product innovations early. RockFlow launched Trading GPT in 2023 — a personalized feed of trading opportunities — making us the first in the world to apply large models directly to the trading opportunity scenario. It performed well after launch, and by September that year, the team was already discussing what came next. We did a series of boundary explorations, and the hallucination problems were severe. We debated whether to immediately push forward with deeper AI product innovation. I was decisive: "Start now." Because any 2C company should be product-driven. We needed to build the most user-relevant product possible within current technical capabilities, then iterate alongside improvements in underlying technology. Many companies prefer to wait until the technology and market are more certain — that's a valid choice, but we chose to start directly.

I often tell the team: the only solution to uncertainty is to start as early as possible. Take this AI stock trading competition — I'm not certain what the strategy marketplace will ultimately look like, but only by starting will we discover where user needs lie, how to iterate, and how to solve problems. Problems are solved in the doing, not in the thinking.

---

"The Purpose of Entrepreneurship Must Be to Build a Platform Business"

Huxiu: You've done secondary market investing, and now you're an entrepreneur. Do these two identities create tension in your thinking? Investing seeks low input and high returns with relatively flexible timelines, while entrepreneurship is longer-term with longer return cycles.

Vakee: Whether it's secondary investing, primary investing, or entrepreneurship, the foundational core competency is the same: reasoning through an industry. The difference is purely in execution. Investing just requires making decisions and pressing the buy or sell button. Entrepreneurship demands strong execution — a willingness to do the tedious work, over and over. In summary: entrepreneurship requires both exceptional strategic judgment and exceptional organizational execution.

During these years of entrepreneurship, I've done two things: after GPT-3.5 came out, I continuously added to my NVIDIA position, and at the end of 2023, I went heavy into Robinhood — at around $8 a share at the time.

Huxiu: Where does this worldview and know-how come from?

Vakee: From work. It's really about abstract reasoning. Entrepreneurship means constantly thinking about strategy — studying competitors and deconstructing the industry every single day. Starting a company without knowing how to do research is terrifying — you have no idea where you're going. The five-year plan I wrote on day one of this company, I'm still using today.

Huxiu: Will these investment philosophies and principles be incorporated into Bobby? Your own worldview and values, for instance?

Vakee: Not entirely. Bobby is more about transmitting objective trading discipline — risk control, stop-losses, profit-taking — rather than my subjective worldview. Subjective elements should be expressed by each user themselves. Users can tell Bobby their own investment preferences. If Bobby can just help users execute the trading discipline and risk control that most people easily neglect — stop-losses, profit-taking — it will already outperform 99% of people.

Huxiu: What was in that five-year plan you wrote on day one?

Vakee: The five-year plan only mentioned building an "all-in-one AI-native trading platform." Now we're considering and preparing more AI-native financial services. We'll have more than just Bobby for trading — we'll have AI for wealth management, insurance, and AI-native digital banking.

We are the financial institution itself. We won't be a service provider to other financial institutions like previous fintech companies. We ourselves will become the AI-era bank, insurance company, wealth management firm — a full-licensed fintech group. Like Steve Jobs wouldn't make phone solutions for other companies; he created and redefined the "smartphone" category itself.

Huxiu: Your philosophy of "going slow" seems different from many AI-native companies today. Everyone talks about rapid iteration, iterating several times a month. Can this logic be transplanted to financial services?

Vakee: If I were just building a research assistant, I could launch in weeks. But users want a closed loop from research to execution. Research analysis alone, with forgotten execution or complex operations — that's useless. The core capability of an Agent is the closed loop; it must have execution power. And to execute financial transactions, you must first become a financial institution. Only then can your Agent complete the specific trading closed loop.

Think of it this way: what you can build in two weeks, others can build too. But if something takes your team, even with extremely high talent density, three or four years of all-in effort — your mindset becomes very steady. The attrition rate in this industry is extraordinarily high; from starting to giving up happens very fast.

Innovation in financial services cannot come from price wars, because price wars mean no profit, no profit means no top talent, and without sustained influx of top talent, there's no sustained product innovation capability.

For a technology innovation company, the real flywheel isn't just a data flywheel — it's that good philosophy and good brand attract excellent people, extremely strong people create excellent product experiences, excellent product experiences reinforce brand distinctiveness, which attracts even better people. This is the same logic as Apple. This is the beauty of entrepreneurship.

Huxiu: What do you see as the biggest change and opportunity that the large model era brings to financial services?

Vakee: I feel like I'm seeing a Steve Jobs era. Financial services will definitely have an iPhone moment. In the future, for services like insurance, payments, and banking, there will be a Bobby for each — I can just ask Bobby to pay my electricity bill instead of searching for a mini-program or digging through an app. Completely more convenient ways to accomplish things, and this opportunity will inevitably give birth to extremely innovative products.

Apple is the most typical example — it brought disruptive change and led the entire smartphone era. Financial services is at that exact inflection point right now.

Huxiu: Have you imagined who future competitors might be?

Vakee: Anyone is possible — after all, this is massive demand, and many people will want to enter. But whoever succeeds will definitely be a new company.

Take Robinhood — it's absolutely number one among trading platforms for the younger generation, and it's done very well. But when it comes to AI-native trading platforms or fintech companies, this field has only just begun.

Huxiu: Robinhood is a mobile internet-era product. Could it transform itself into a strong competitor in the AI-native finance space?

Vakee: Robinhood is already very strong; it doesn't need to transform to remain strong. But there's a special dynamic in financial services: once a company scales up and its profits stabilize, it becomes less willing to pursue radical innovation. I dared to build Bobby because of the opportunity of this era, and because I'm willing to go all-in for several years to realize it.

Huxiu: Would a new AI-native company entering the trading platform space today pose new challenges and threats?

Vakee: Fintech entrepreneurship requires waiting through cycles. You need to first spend years applying for various financial licenses, building trading systems, doing meticulous compliance preparation, connecting with all upstream and downstream financial institutions, and building brand reputation. This is slow work. Not every elite team is willing to endure several years of doing the hard but right thing.

Huxiu: Have you thought about what might cause RockFlow to fail in the future?

Vakee: Compliance — because global financial compliance is extraordinarily difficult, and we deeply respect that. But objectively speaking, our company won't "die" at this point, because we're already profitable. In the worst case, we could become a very profitable large-model quantitative trading team — that's already my core area of expertise. I've told the team that at any point we could become the "Renaissance Technologies" of the large-model era, but that's not my goal. The purpose of entrepreneurship is definitely to build a platform business, to do the most challenging things.

For me, failing to seize an iPhone moment-level opportunity for financial services would itself be failure.

---

Huxiu: What do you think an AI-native organization looks like? Is RockFlow practicing this?

Vakee: I think RockFlow is a very typical AI-native organization. Our human efficiency is exceptionally high. Theoretically, for our current business scale, other financial institutions would need 20 times as many people. Although our business is quite complex — combining financial institutions with technology companies — we've achieved extremely high human efficiency. The core reason is that AI makes excellent people even more excellent.

Huxiu: Do you provide each person with customized AI tools to improve efficiency?

Vakee: Everyone uses various AI capabilities and AI tools to solve problems that previously required different approaches. For example, writing code and web development can now be done with AI — one person can build an app. Product design can use AI to generate proposals. Growth operations can be automated with AI.

Huxiu: So when hiring, how do you filter for people who fit this AI-native organization?

Vakee: The first criterion is passion for investing and finance. Second, they must be an AI master — this is easy to verify, just have them demonstrate on the spot. For example, I ask them to build a simple application, and they can produce it immediately. When encountering something they haven't dealt with before, they'll immediately turn to AI for help, asking AI "how should I think about this problem."

Huxiu: What level is the company's ARR at now?

Vakee: I don't think discussing ARR itself is very meaningful. What matters for an impressive company is human efficiency. If you have only 100 people but can earn profits comparable to Taobao — that's impressive. Of course, having 100,000 or even 1 million people is impressive by another definition, but it's not what I define or pursue. I only care about human efficiency ratio and profit margin — using the fewest people to earn the most profit. That's my standard for judging an excellent company. In the brokerage industry, inflating ARR is easy — constantly opening accounts with free phone giveaways, cash rewards for five trades, pulling in loads of users with extremely high marketing costs who don't stick around. That means nothing. Our industry shouldn't go down that path.

Huxiu: What's the average age of people in the company?

Vakee: 28. Post-95s and Gen Z make up an increasingly large share. We only recruit the absolute top people; everything else gets replaced by AI.

Huxiu: How do you attract these top people to a company that's still in the startup phase?

Vakee: I don't think it's about deliberate tactics — it's about fit. You just need to express yourself authentically. If you click, you naturally come together; if you don't, you don't. A lot of people might find my ideas odd, and that's totally normal. But sincerity is always the ultimate move. The more honestly you express what you think, the more you'll attract people who are on your wavelength.

Huxiu: You went after overseas markets from day one, which really tests a team's understanding of those markets. Is there any experience you can share with other founders?

Vakee: Honestly, there's nothing to worry about — just go do it. We have offices in four countries and regions, and we're planning to open Silicon Valley and London offices next year. The international capabilities of post-90s and Gen Z Chinese teams are already incredibly strong.

Huxiu: From investing to founding a company, has there been anyone who's had a major influence on you, or anyone you actively study?

Vakee: There are relatively few people who can directly shape you in day-to-day work. But I study the classic ideas and public statements of outstanding figures — people like Munger, Steve Jobs, Warren Buffett, Elon Musk, and the founder of Telegram. You'll notice that excellent organizations and founders share a lot of commonalities, but also have differentiated aspects worth learning from. It's hard to say you'd completely copy any one person. Just like investing — you can't copy someone else's investment style.

How a person invests or builds a company is shaped by their worldview, values, and outlook. The same goes for me — my company is built on my own worldview. There's no right or wrong; what matters is that you can stand behind it, stick with it, and find an approach that fits. Your own approach might not be accepted by most people, but that doesn't affect the outcome. Look at Elon Musk — his way of doing things might not fly in Eastern culture, but he's still crushing it.

In Conversation with RockFlow Founder Vakee: Killing Every App, From Stock Market Prodigy to AI Gambler | BlueRun Ventures Family Headlines

BlueRun Ventures Continues to Double Down, RockFlow Closes New Tens of Millions of Dollars in Funding | BlueRun Ventures Family

Genspark's Kun Jing: Dancing with Microsoft Agent 365, Making AI Agents Everyone's Work Partner