Don't underestimate everything you see in life.
If there were a "24/7 online investment partner" that, through natural language conversation alone, could help you analyze real-time market dynamics, generate personalized AI investment portfolios based on life inspirations and work insights, complete stock/options trades with one click, and even offer rational strategies when you're anxious—how would such an AI Agent redefine the way we "participate in the world of investment"?
We invited Singapore fintech company RockFlowFounder and CEO Vakee Lai to talk about the financial trading AI Agent she just launched—Bobby. This is not just the birth of a product, but also the concrete realization of the mission to "make investing simpler": from starting to trade stocks at age 9, to Baidu Fengchao product R&D, to early-stage AI investor, and now to entrepreneur "reshaping the investment experience with AI," every step Vakee takes answers the question: how can ordinary people, in the most natural way, turn life insights into investment actions?
We will unfold this from three dimensions:
1) Bobby's "not simple" story: Why did RockFlow spend two years polishing this Agent? How does it solve users' pain points of "high threshold for investment awareness" and "complex tool operations"? How do real cases, such as a UK user using Bobby to watch the market and buy Circle stock for a double, validate the value of an "AI-native investment experience"?
2) Vakee's "investment worldview": From the cognitive advantage of Fengchao's advertising monetization to the "amplified cognitive returns" of a 15-year investment in NVIDIA, how does she use the philosophy of "earning money within your cognition" to traverse both primary and secondary markets? As a rare female CEO in the tech world, how does she balance entrepreneurial challenges with the mission of "letting investing fly into ordinary homes"?
3) The "rock spirit" of future investing: Why is saying "remove all GUI and keep only Bobby" not radical but inevitable? When the post-00s generation expresses attitudes through investing and conveys views through trading, how does RockFlow use AI Agents to create a "lifestyle investment playground"?
Finally, Vakee will combine the Xiaohongshu viral article "How Ordinary People Can Discover the Next Trend Around Them" to share her exclusive perspective on capturing investment opportunities from life and work—perhaps the starting point of the next "cognition monetization" is hidden in your daily life.
👦🏻 Koji
At this week's Crossroads, we invited Vakee. Vakee is the founder and CEO of RockFlow, and RockFlow recently launched its own AI Agent product called Bobby.
Today we hope Vakee will talk with us about her own story of starting to invest at age 9—first as a young investor, then working in primary markets and secondary markets, and now her startup RockFlow aims to help more young people enjoy the fun of investing and also gain investment returns.
Vakee has actually also been on the Crossroads podcast, about a year ago, but at that time what we talked about was not the RockFlow product; the title of our conversation was"Are My Investor Friends Really All Clearing Out NVIDIA?"
At that time we talked about whether NVIDIA stock was worth buying, and I remember Vakee firmly told everyone to keep building positions, right? If you had listened to Vakee's advice then, you should have already gained pretty good returns by now.
So please let Vakee say hello to everyone again, and also briefly introduce herself, introduce RockFlow and your latest AI Agent Bobby.
👩🏻 Vakee
Okay, hello everyone, very happy to be at Crossroads again.
I'm Vakee, founder of RockFlow. Today I've brought our new AI Agent product, Bobby, and I also hope to share more of the RockFlow story and the story of Bobby's birth with everyone.
What is a trading Agent, and how do you use it?
👦🏻 Koji
Please introduce what Bobby is. Because Bobby feels like a recent major focus for you, and an AI Agent product you launched after building up a lot of momentum.
👩🏻 Vakee
Bobby is the first AI Agent in the field of financial trading.
Simply put, it's your 7 × 24-hour investment partner. Its biggest feature is that it can complete your entire investment loop, from daily inspiration discovery to investment analysis and research, to generating trading strategies, to order execution, and even including your position management after execution. It's like having your own hedge fund team.
There are traders, analysts, and a risk management team, which can help you better realize your ideas in investing.
This is a very vertical field, and it's hard to do, but we still managed to realize such a product that can close the loop.
👦🏻 Koji
What is the biggest difference between it and the previous RockFlow App?
👩🏻 Vakee
The previous RockFlow was still in the form of an App. We wanted it to be simple enough to let everyone complete trades very conveniently. But Bobby's entire form is natural language dialogue. We actually hope that later there may be "no more RockFlow App." The main interface is just Bobby, and you only need to talk with it and then complete the entiretrading loop。
I used to tell everyone that RockFlow is clean in design and simple to use. These meet the needs of some people, but in fact they are not the most fundamental differentiation. Now, with Bobby, this Agent-based interactive application form, I think it is a groundbreaking product that can solve problems in all complex scenarios.
After everyone uses Bobby, they won't ask me anymore, "What's the difference between you and Robinhood?"
👦🏻 Koji
Speaking of "there may be no more RockFlow App in the future," it just so happened that yesterday was the first day of Sam Altman's sharing at Y Combinator Startup School, and one very important point in it was that "UI interfaces will disappear in the future."
Actually, same as what you just mentioned, are you really making this plan that extremely now? When do you plan to kill the App and keep only the Bobby Agent?
👩🏻 Vakee
The plan is to launch a product with only Bobby this year.
Koji and I have both been product managers. In the past, when we made products, we had to seek the greatest common denominator, so we would extract from the target user group the feature points and demand points that we thought were most important. That was the logic of making products before.
But we clearly found that in complex scenarios, feature-based interfaces are actually very difficult to use to fulfill different people's needs — for example, this particularly complex scenario. Actually, stepping back, even OTA Apps such as Ctrip, or even Baidu Maps, will unconsciously make the App super complex.
Let me give a few examples. With Ctrip, I often book hotels, but I simply can't choose or filter out what I want, because its filtering dimensions are different from mine. The dimensions I want cannot be filtered all at once. This is actually quite absurd, because in fact these dimensions, this data, and these tags all exist in the database. Instead, it is the App form, or this kind of page based on limited features, that seriously hinders me from doing what I want to do.
I think in the future it will be very simple: Agent plus database.
My genuine feeling—for example, with Bobby, we used to have to develop many order placement methods. In fact, our requirements pool was especially large, but every time everyone had to debate for a long time about whether to first build trailing take-profit orders, or first build that order. Because the development cost of each order is not low, from frontend to backend, from algorithms to engineering, and its occupation of the interface is also very strong.
We all know that doing subtraction in products is very hard, while doing addition is the easiest, and then there will be 10,000 features in your App.
👦🏻 Koji
Yes, especially for a brokerage APP like RockFlow, it is especially easy for it to turn into having 10,000 features.
👩🏻 Vakee
Exactly, so what problem is Agent actually solving? In this kind of complex scenario, every requirement is actually quite unique.
For example, Bobby, in the future we will no longer develop any such detailed features in the App. Users can directly tell Bobby what I want to do, and Bobby can help you implement conditional orders that have never appeared in this world, because it can reassemble those features in an AI way.
These forms of conditional orders do not exist on any existing internet brokerage App.
It is impossible for one [App] to provide 80 ways of placing conditional orders; there must be trade-offs, but in fact users may indeed have 800 different ways of placing orders.
So in complex scenarios, interacting with Agent using natural language can better fulfill each user's personalized needs.
AI Agent is a completely new experience. It will change the way product and research work is done, and even change the appearance of the App—we will see a huge paradigm shift.
For example, my mom yesterday asked me to book flight tickets for her. She and my dad are going to Urumqi for a trip. They cannot buy too late a flight, and then it is best not to have a layover, or the layover time should not be too short. There is no way for me to interact with the Ctrip App to express this matter.
👦🏻 Koji
So you are just your parents' Agent. You are helping them interpret requirements, break down tasks, and helping them complete the order placement, closing the loop.
👩🏻 Vakee
Exactly, but in reality, look at Ctrip—what I want to express is all in its database, but the App can solve nothing. Baidu Maps is the same, and Amap is the same. For example, I am in Xierqi and want to arrange with a friend in Shuangjing to find a place to eat hot pot. I want to tell him, Baidu Maps, can you find a place for the two of us that is convenient for both of us by subway, and then a non-spicy hot pot. But it cannot be searched on the App, so it is very hard to use.
So I think the change I see is this: all relatively complex scenarios will definitely be database plus Agent in the future. I think Apps based on feature arrangement will not exist in the future.
👦🏻 Koji
This year there is a wave of enthusiasm for Agent.
At the beginning of the year, in the first episode of Crossroads, we invited Dai Yusen, managing partner of ZhenFund. At that time we did a New Year conversation, and the title of that episode was The First Year of Agent。
At that time, actually mentioning the first year of Agent was not yet so obvious, but at the time we did indeed see some signals, especially the release of Devin, which made us feel the maturity of Agent technology and the innovation of the interaction paradigm.
Manus and Lovart were released, Crossroads published in-depth reviews within 24 hours; AI Agent products such as Flowith, Clacky, and today's Bobby also chose to launch at Crossroads.
I want to ask Vakee, when did you start planning to make the Agent product?
👩🏻 Vakee
We just happened to look into this internally these past couple of days. Regarding the Bobby product, it was initially called Rockbot, and work on it actually started in April 2023.
The first-level page feature in the current APP is called AI Strategy. It was originally called Trading GPT, and it was also the world's first to directly use GPT to achieve the entire process from what happened to what stocks or options you can buy, and it was a personalized experience—real-time push of personalized trading opportunities.
After we launched this product, we immediately began internal discussions about what product form we should use to present the next step in the AI-native experience. Then yesterday I even found that demo video. At the time, when we were discussing this so-called AI native product form, the term Agent did not exist yet.
👦🏻 Koji
It was May two years ago, right?
👩🏻 Vakee
May two years ago, May 2023. At that time, our exploration was that we had launched Trading GPT, explored the product form, and based on this bot and Copilot product form, had already made a very complete demo. Then in September 2023, we officially established the project internally and began working on what is now Bobby. Because our internal documentation culture is especially good, all meetings and everything have very clear records.
The first point mentioned was Agent. So actually, in September 2023, we decided to use this current Bobby form, using Agent's basic architecture, to realize the AI product experience we wanted. I also looked at the meeting minutes from that time. The first part of our discussion was the current boundaries of AI capabilities, the capability boundaries in 2023 at that time: what could be done, what it did well, and what it still could not do.
👦🏻 Koji
I'm quite curious: when you established the project in 2023, what new breakthroughs are there between Agent's capability boundaries then and the capability boundaries for making Bobby today?
👩🏻 Vakee
At that time, we actually saw that Agent could already support logical reasoning and multi-stage contextual thinking, and could automatically choose what to do next based on the current state. Calling APIs was no problem, and the ability to access and maintain a knowledge base was also ready. These capabilities were OK. The biggest problem at the time washallucinationwas very serious. At that time I needed to make a decision, because there were actually different voices within the company. Seeing such serious hallucinations, people wondered whether we should wait a while longer, until several models and foundational capabilities were more stable, before doing it.
But at the time, the main colleagues and I were actually still very firm that we definitely had to explore first. This is the so-called difference between a "company that builds ships" and a "company that builds towers." If you are a "company that builds ships," then as the base models and underlying capabilities improve, your product capabilities will improve along with them, while building towers will get flooded.
So first we judged whether the product we wanted to build was "building a ship or building a tower." Building an application in a field with very high industry barriers is definitely the "building a ship" route. Second, if I am that kind of company, should I wait? Because we had already been working on RockFlow for one or two years at that time, I knew deeply that the financial industry has a large amount of knowhow in products and engineering. If I want to build a vertical application in this field, the optimal solution is to do it quickly, because in the process you will solve many very detailed problems.
👦🏻 Koji
To build a good financial tool, there are still many engineering tasks and details. If you do it earlier and more meticulously, you have the advantage of leading the race. I also feel this way when using RockFlow and Bobby myself.
After Manus was released, six teams within ByteDance immediately appeared to copy Manus. Today, indeed many companies have already produced high imitation replicas, but these replicas have not shaken Manus's position — Manus is not as easy to 'copy' as everyone imagined.
But when I look at Bobby, I feel this is even more impossible to 'copy' — at least you won't be able to copy a Bobby within half a year. Behind this there is a large amount of trivial, delicate work to be done.
👩🏻 Vakee
For Manus, based on my superficial understanding (what I say may not be right, because I am indeed an outsider), although it is a general-purpose Agent, its entire interactive experience contains a lot of understanding of browsers and engineering understanding at the interaction level. This team must have core members with very strong prior accumulation.
Some things are common. For example, as just mentioned, if I want to build Bobby, you first have to be a team that can do financial engineering. For example, handling corporate actions may have hundreds of possibilities. To this day, there is no vendor in the world that can help you solve these detailed problems well, including IB, including Futu, including us — we all handle them step by step ourselves. So why is Bloomberg, including similar domestic companies like Wind, still very valuable? Because financial data is all hard work.
In the financial industry, the requirements for data accuracy are very high, and tolerance is very low. You have to do it very meticulously. So your financial engineering team must be built by doing this business. Whether we are making general or vertical Agents, there are still many, many details in the product development and engineering problems to be solved in the field we are in, and I think this is the barrier.
So RockFlow did not do Bobby in its first year. We were founded in 2021, first spent a few years as a US stock brokerage, and built a full set of counter trading systems. With users in various countries, we understood the details of needs together with users, and only then began to make the product Bobby, still with the original team. In addition to AI R&D capabilities, vertical Agent applications must also have product and engineering teams with a deep understanding of the field. In fields like finance and healthcare, the team's cognition of the industry will all be reflected in the product.
👩🏻 Ronghui
What is Bobby's weight within the company? At the time, did you treat it as an experimental product, or as a bet on the future?
👩🏻 Vakee
At first it was a priority project, and then especially this year it became absolutely all in. Throughout the R&D process, this quarter's and last quarter's OKRs, except for requirements related to trading stability, all gave way to Bobby. You
asked a particularly good question, and this can also answer, for example, why predecessor peer companies do not do it, and how they would treat this matter. The core is that we are doing it all in, treating it as the highest priority, and we are doing it as if there would be no RockFlow App in the future. If you only treat it as a feature, it is very hard to succeed.
When I used to do VC, I often asked founders what to do if giants also do it. In recent years I have deeply felt that no matter how big the team is, the ones who can really fight are only those ten or twenty people. Any PK between companies, or competition between industries, depends on who dares to go all in with these 20 people on this project.
If you dare, then I think everyone has a chance. If you do not dare, and you treat it as internal innovation, then it is very difficult. Because startup teams fight with their lives, while big companies do internal innovation, doing it with million-yuan salaries, completely different fighting states.
👩🏻 Ronghui
How much of this comes from your certainty that Bobby's Agent product form will replace RockFlow in the future? How much comes from the confidence given by the industry?
👩🏻 Vakee
I think mainly 99% is the former.
At the intersection, I interviewed many founders of AI Agent products, but I barely have the energy to care about what others are doing. Many people ask me, what do you think of Manus? What about Flowith? Actually, I can't pay attention to too much. You really put almost all your energy into charging forward for the goal you believe in, because this thing is too new,Bobby is the world's first thing of this kind。
We fantasized a future for ourselves, and then we went all in charging toward that future. Our team is 80 to 90 percent INTJ, so everyone is very J. That thing, I just have to do it, and I have to finish it as soon as possible.
Now I can clearly feel that the world is undergoing huge changes, and I saw that Agent plus database is the product form of the future. What I need to do is use this form to satisfy user needs in this field. I firmly believe this, and I myself am now using Bobby for all trading and order placement. It is now more convenient and easier to use.
Since I was little, I have never had peer pressure, and I never cared what others were doing. Each generation has its own needs. ToC entrepreneurship is about whether you can make a better product that can satisfy this generation's needs. So in this process, it is more about thinking about your users and thinking about your team, what your users' needs are, and whether your team's execution efficiency is high enough. That is enough.
👦🏻 Koji
This sounds like a pretty happy kind of entrepreneurship. INTJ can polish one thing to death, very exquisitely.
Bobby has actually been in beta for a while, and there should have been several hundred users in the previous beta. I want to know if there are any fun stories?
👩🏻 Vakee
Because we have been gradually opening it up over about these two weeks, I feel quite excited. Bobby now releases a version every day and iterates the model once a day. This is what we can do. Even though the RockFlow App has come this far, there will still be all kinds of small problems, so we just keep iterating quickly.
Let me give a few fun examples. The first is the day Circle went public. We had a user in the UK. This was the first stock she bought, and also her first trade on RockFlow, because everyone discussed Circle in the group, and then she directly told Bobby, after it goes public, place this order for me, buy this much for me, and then she went off to do other things. The next day she found that Bobby had bought in at 80 dollars and had also placed a stop-loss order for her. This stock has now doubled. She shared the complete chat record with Bobby and a screenshot of the returns in the group. She felt that without Bobby, the time when she started investing in US stocks might have been delayed.
Actually, he had already been a RockFlow user for quite a while, but he had never bought anything and did not know how to start. Bobby actually has many boundaries to explore. I even previously did not know that Bobby could one-click close all positions. We have always had a need for one-click closing, but it was never done and never made it onto the schedule. But now we found that users figured it out themselves, and Bobby completed it for him.
👦🏻 Koji
So will Bobby have some position-closing strategies?
👩🏻 Vakee
It will make some adjustments. For example, if you have 1 million US dollars of Tesla stock to close all at once, it will help you place orders in batches and then smooth out your slippage.
👦🏻 Koji
I myself used Bobby to try a very, very beginner-level instruction. I deposited 3,000 US dollars, and then told Bobby: “I now have 3,000 US dollars, help me buy stocks, I want to make money.”
At that moment, I really wanted to give it a try. If I blindly trusted Bobby, what would it do for me? In the end, it bought me 6 stocks, all very stable, including two index funds (SPY and QQQ), and it also bought Apple, Nvidia, and Coca-Cola. This is actually very beginner-friendly. When you have $3,000 and don't know what to buy, handing it over to Bobby will most likely give you a better result than impulsively placing an order yourself.
👩🏻 Ronghui
A question also came to mind here. For many beginner users, like people who don't trade stocks much, such as myself, it sounds very simple to use. I asked a lot of questions, and with the same prompt, I also gave it to several other Agent tools. My feeling is, first, Bobby is very simple to use; second, it is helping me operate something with a very high barrier to entry. So for users like me, how would you explain how Bobby serves them better and how it helps me achieve my goal?
👩🏻 Vakee
The example just mentioned is actually a beginner user who has never traded. Investing has two barriers: one is your psychological barrier, where you think it is very difficult; the second is the operational barrier, which is troublesome and you really don't know how to do it. Bobby, as this kind of Agent with natural language dialogue, solves users' troubles at both levels.
Bobby lets you express what you see and think of in life, turning an idea into a trading opportunity. This interaction form inherently lowers your psychological barrier. When you talk to an existing brokerage app, you don't know how to communicate with it. Bobby is a bit like you communicating with a person in a more natural way. It lowers your psychological pressure and barrier.
👦🏻 Koji
But actually, when users chat with ChatGPT or Yuanbao, the replies they get may be the same too?
👩🏻 Vakee
At the interaction level of natural language dialogue, it is the same. Using this form first helps lower your psychological barrier.
But more importantly, many people find the trading scenario relatively complicated to operate. The biggest difference between Bobby and other large-model chatbots is that Bobby helps you close the loop on the act of investing and trading. For example, at the beginning of the year, if you ask Bobby, "Labubu is really popular, what should I do?" Bobby might say you could consider Pop Mart's stock. Or if it has already become too popular now, Bobby will remind you that you should be cautious about investing in Pop Mart. It will give you an investment strategy that corresponds to your inspiration, combined with the market environment and your risk preference, and it can help you complete the order placement.
I think the most important thing in building a vertical AI Agent is to help users get this thing done.
👩🏻 Ronghui
Bobby actually took two years for the entire development process. Why did it take so long? Did you take any detours along the way?
👩🏻 Vakee
We officially started the project in September 2023. It has been a year and a half, and basically there were no detours. To build a vertical Agent and achieve a closed loop really takes a lot of time, and a lot of details need to be polished. In the earliest versions, we did not consider speed too much. But later we found that being slow is the original sin. If you're slow, there's no waiting; users aren't that patient.
👩🏻 Ronghui
What is the standard for measuring slowness? What time counts as slow?
👩🏻 Vakee
We have actually done a lot of optimization on the matter of "slowness." In the early days, Bobby's response speed was indeed not fast enough. A round of dialogue might take ten or twenty seconds to produce a result, which is completely unacceptable in a trading scenario. Later, we redesigned the entire Agent scheduling system, optimized the orchestration capability, and raised both efficiency and performance. Because I have always felt that speed is the first hurdle. Only when you are "fast enough and stable enough" can you even talk about "productization."
But if you want to make it a product, or even further commercialize it, you still have to face a very critical issue: controllable costs. You can't consume a large number of tokens on every interaction, and you can't let users drop out of the process because they've waited too long. So when building an Agent or a large model application, these two problems essentially always have to be solved:First, speed and experience; second, cost control。
Internally, we have actually put a lot of effort into these "structural problems behind the features." For example, how do you design the architecture so that token consumption stays stable? How do you ensure that every time a user asks a question, they can get an answer within 1-5 seconds? In fact, these underlying mechanisms are often harder than the surface features,but it is exactly these that form the foundation for an Agent product to be deployable.
👦🏻 Koji
Among these beta users, what data metric are you most focused on? For example, how many rounds were chatted? Or how many orders were placed? Or how much money it helped him make?
👩🏻 Vakee
Activity metrics, usage duration. You have to have multiple rounds of conversation with it, and then it has to actually help you, so it must have solved some problems or doubts in the process.
👦🏻 Koji
Just now you shared the case of Bobby buying the stock Circle. It sounds like Bobby placed the order in the middle of the night? Was there any user confirmation at the time? I think with automatic order placement like this, if it helps users make money, of course they're happy; but if it loses money, will the blame fall on Bobby?
👩🏻 Vakee
All orders are confirmed. For example, "how much to buy," Bobby will repeatedly confirm, at what price to buy, and only after confirmation through dialogue will it execute the instruction. So it is still reliable.
👦🏻 Koji
Are there any stories about people using Bobby to "make money they might not have been able to make before"?
👩🏻 Vakee
That day Trump and Musk were arguing, and a user who trades Tesla very heavily asked Bobby, what should he do now? Bobby said that since the two of them are arguing, Tesla will most likely keep falling, and he could consider buying Little Bear, which is one of our put options. Then after multiple rounds of dialogue, he chose a put in the conversation, and this user bought this Little Bear, and then became the No. 1 on that day's hotshot leaderboard, selling after a rise of more than 400%. Then he also gave Bobby the largest tip. He even said that if it could be set as a "profit-sharing tip," for example 1% or 5% of his profit all going to Bobby, he would be willing.
He also shared several feedback points in the community:
First,he placed this order entirely because of Bobby's conversation. It was not a judgment he made out of thin air, but rather Bobby structured the cognition and turned it into a clear and executable decision.Of course, in the end, placing the order was his own decision, but Bobby played a very key triggering role.
Second, he said Bobby was the first product where he truly felt "understood" when using AI for trading. He has used many products, from IB and Futu to Robinhood, and even some AI tools, but none had this combined experience of "conversation + understanding + customization."
When he uses Bobby, he can very clearly feel that this Agent is analyzing his past trading behavior, understanding how he does risk control and how he selects targets, and then all the suggestions hit exactly what is in his "heart." It is like an investment assistant that truly understands his rhythm and habits, rather than a stiff trading tool.
This is typical feedback from an experienced user. Previously, many people felt that RockFlow was too "lightweight," too simple, and not professional enough. But after they started using Bobby, they found that it is actually the tool that best supports complex decisions right now. I think this comes back to why we shifted from a traditional feature-based app to a conversation-centric Agent form. This form naturally broadens the range of users you can serve. From complete beginners to seasoned traders, everyone can get something from Bobby that matches their needs.
These two examples just now are two very different experiences, and each of them felt that Bobby, as an Agent, can better cover people with different needs. From a feature-based app to the current Agent conversation form, the range of users that can be covered has been expanded all at once. In the past, some more experienced users felt RockFlow was too simple, and many features couldn't meet their requirements. But now mature traders also use Bobby and place orders with it, getting their first positive feedback.
👦🏻 Koji
Right now Bobby still needs users to proactively ask a question before it responds. Have you considered letting Bobby proactively give users some suggestions?
👩🏻 Vakee
Our next version of Bobby will proactively give you things you might want to know. There will be some trading inspiration, but it won't suddenly jump out and say you should buy this.
👦🏻 Koji
What kind of trading inspiration?
👩🏻 Vakee
For example, macro information like war, or some stocks related to your holdings. For example, if you're very interested in stablecoins, it will push this related information to you. We also found that everyone's most essential need is to call him, to let Bobby call users.
👦🏻 Koji
Users want to be reminded?
👩🏻 Vakee
Yes, he just wants to be reminded.
👦🏻 Koji
I have a friend who once managed a family office in Hong Kong for a certain big shot. The job itself was quite interesting—a small team of four or five people serving a man in his nineties who still keeps a close eye on market trends every day and calls them countless times.
The calls roughly fall into two categories: one is clear trading instructions, and the other is vague research requests, such as "help me check whether there's any movement in this sector recently" or "is this ticker worth adding to the position." It sounds like their team was like the old man's "Bobby"—an always-online AI Agent, except a real-person version. Everyone on their team had a strong background, basically MBAs from prestigious schools and senior practitioners in the financial industry, with annual salaries of several million without question.
But I'm thinking these "senior wage earners in the financial industry" will probably soon have to compete with the Bobbys.
👩🏻 Ronghui
I was just about to add a question. During the roughly year-and-a-half R&D process, you mentioned some very important decisions, such as how to save tokens and save costs. I'm actually quite curious. Looking back, which decisions do you now think were correct or very meaningful? This might be quite instructive for many entrepreneurs.
👩🏻 Vakee
The first question every entrepreneur should ask themselves is whether you want to build a vertical or a general-purpose Agent, and why you chose that approach. The reason we chose vertical is actually very simple, because there are only two kinds of needs in this world. One kind is where people just don't want to do it, and whether it's done well or badly doesn't have that big an impact—it's not fatal. For example, some scenarios at work: help me write an email, analyze a travel route, generate a research document. I may not want to spend a lot of time on these things myself; if I spend the time writing it myself, I can get it to 70 points. If an Agent helps me and maybe gets it to 60 or 65 points, I can accept that too. This is most of the needs in life.
The other category, such as some precision scenarios in finance, healthcare, and manufacturing, is different. Getting to 70 points versus not getting to 70 points may be fatal, and the difference can be enormous. So in this type of scenario with high fault-tolerance requirements, the standard must be met. Using a workflow-based approach can relatively quickly and well reach the 70-point standard.
So you have to distinguish which industry you are in, and which type of problem the vertical you are solving is actually solving. This determines what form your architecture should be based on—this is the first business question. The second is the optimization question, which is actually solved during the process. Over the past year or so of development, we have gone through multiple iterations. The core idea is to choose the best-practice solution around the needs of specific scenarios.
Vakee's personal experience: investment, Baidu, AI
👩🏻 Ronghui
Could you talk about your personal experience? Especially the story of starting to trade stocks at age 9—when I saw related reports some time ago, I was very curious about that experience.
👩🏻 Vakee
This was mainly influenced by the family environment. In my memory, stock trading in the 1990s was a nationwide phenomenon: TV broadcast stock market quotes from 1 to 3, K-line charts in newspapers, and orders placed by phone. I often went to the big-client rooms at securities brokerage offices.
👩🏻 Ronghui
Did this environment naturally expose you to a large amount of investment information and lower the barrier to entry?
👩🏻 Vakee
Exactly. Later, doing US stock trading was a similar situation—after I returned to China in 2013 and joined Baidu, the colleagues around me were all trading China concept stocks. Internet practitioners were among the earliest groups in China to come into contact with US stock investing.
👩🏻 Ronghui
Please outline your career trajectory. Has there been anything related to investment trading and AI?
👩🏻 Vakee
My career experience covers product and research at a big company (Baidu Fengchao), the primary market (Baidu Investment Department's technology track + VC), secondary market investing, and now entrepreneurship. Apart from never having worked at a state-owned enterprise, I have done all kinds of professional roles. In 2013, I gave up a sovereign fund and chose to return to China precisely because I recognized the mobile internet wave. Before starting my business, I was either doing AI or investing in AI, and I had always been investing in US stocks.
👩🏻 Ronghui
You said before why you started RockFlow, and the inspiration came from "retail investors vs. Wall Street." So when you talk about "belief," your entire entrepreneurial process is actually executing the idea of "making trading simpler." Can you talk about how you felt when you saw "retail investors vs. Wall Street" at the time?
👩🏻 Vakee
It was a beginning. Because it was a phenomenon-level event, it prompted you to start paying attention to this group of people and this matter. At that time I had actually already returned to the secondary market, and I began to pay attention to the changes happening in the trading world—first of all, the way young people invest had undergone tremendous changes. In the past, the goal of investing was to make money, but for Gen Z, investing not only carries making money but also other attributes: they want to express their values.
Let me give the simplest example: in the past, value investing, like Warren Buffett's approach, looked at a company's value from the perspective of investment research, for example, whether the PE ratio was reasonable; but what young people today call "value investing" is "if I think it has value, then it has value." Investing has become a completely different experience. GameStop is a typical case. Some people said, that's where I bought game discs when I was a kid, those are my memories, it shouldn't be shorted by you guys. To him, this company has value in his life, so he uses investment behavior to express his attitude. Investing has become part of a lifestyle.
Because of this event, I started researching this group and found that young people all over the world want to buy Tesla and Apple stocks and are interested in US stocks. But you'll find that, apart from China and the US, almost nowhere else has good US stock brokers; supply is extremely limited and penetration is very low. This is a clear opportunity I saw.
👩🏻 Vakee
Speaking of why I started a business, at the time I had several criteria for judgment:
First, I want to do something big enough. Because I know that no matter what you do, it's hard, so you should do something big. Finance is an industry even bigger than the internet; apart from energy, the biggest is finance.
Second, I must build an AI company. When I was doing investment before, I often said that advertising, finance, and gaming are the three scenarios most suitable for AI to land. Finance especially is a pure data scenario, very suitable for rapid AI iteration, and trading is the most ideal entry point.
Third, I must have passion for it. Because starting a business is too hard. I have invested in nearly 30 startups; many that look glamorous today, or have even gone public, actually had very difficult processes, so you must truly love this thing. If it's not "I must do it," you won't be able to persist. I am also often asked for startup advice by some young students, and my first question is usually: Is this something you must do? If not, then you don't actually have to start a business.
Fourth, I have to be able to do it. This comes back to how I define this goal. If I want to build a trading platform for the AI era, what do I need to do?
At the time I set two keywords for myself:
- All-in-one: one platform where all categories can be traded—US stocks, Hong Kong stocks, stocks from various countries, futures, options, lottery, contracts, forex, Crypto, all can be traded.
- AI Native: I hope to provide a native AI experience.
All-in-one and AI-native are the two core problems I need to solve. If I can do them well, it will be a "new species," an opportunity to provide a completely new experience for users in a new era.
So can my team and I do this? Looking back later, I found that every step of the ten years of work experience before starting the business counted. From doing quantitative work in London, to joining Baidu Fengchao and becoming an AI user, to entering Baidu's investment department and VC investing in AI, from AI product research to investing in AI, plus always being immersed in the secondary market, finally starting RockFlow seems very natural. To find someone who understands both AI and investing and can also tirelessly be hands-on in daily product research and operations, I am one of the very few. To be honest, for many people with similar backgrounds, those who can continuously make money through investing, the vast majority are unwilling to start a toC product business; it's too hard.
Even today, one of the questions people ask me most is: "Vakee, your investment ability is so strong, why do you still go start a business?"
My answer is:This is a mission. I really want to make this happen. From being exposed to investing as a child to truly understanding the logic behind trading, I have always believed that investing is not an unreachable thing—it can be something the public participates in. For example, the reason I started trading US stocks was because I was at Fengchao, where I was exposed to China concept stocks and American internet companies. Fengchao built an advertising system, and 90% of internet companies' profit models are ad monetization, so I naturally had a cognitive advantage in this field. I could even open the Taobao homepage, look at the position and height of the first few ads, and roughly judge whether Alibaba's performance for the quarter could be met—because that was my daily work. Later, when I started investing in AI, I bought Nvidia stock, also by the same logic, amplifying cognitive returns through the secondary market.
What requires extra effort is odds design: for example, should you buy stocks or options? How should you allocate, and how should you protect against risk? These are trading-level techniques. But judging whether 'the direction is right' actually comes from life and work. And this is also why I think: if I can do it, many people can actually do it too. For example, when GPT-3.5 came out, many people already knew Nvidia would benefit, and even knew AI was the future. But many people did not turn this cognition into an investment decision, nor did they act. I have seen too many examples like this. Some people had long been buying things on Pinduoduo every day, but had not bought a single share of the stock. Many people who bought Tesla cars very early had also never invested in Tesla stock.
👦🏻 Koji
I think this experience is especially interesting. Because Vakee recently had a post go viral on Xiaohongshu, titled 'How Ordinary People Can Discover the Next Trend Around Them.' The core she wanted to convey was to encourage everyone: turning observations of trends discovered in life into investment decisions can actually make it possible to obtain real returns.
Just yesterday, Xiao Hong suddenly told me on WeChat that a year ago, when I met him, I said—Pop Mart's best-selling IP was no longer Molly, but Labubu, and it was growing very fast. We both have a regret: at the time, we did not turn this discovery into an investment decision.
👩🏻 Vakee
Exactly. The money was falling from the sky, and you didn't even reach out to catch it.
👦🏻 Koji
So I think now that Bobby exists, I am more likely to catch it!
👩🏻 Vakee
Yes. In fact, whether it is RockFlow or Bobby, essentially what I am doing is using different product forms and new technological means to let people turn everyday cognition into trading opportunities. And the simpler this action becomes, the easier it is to implement. For example, you only need to say one sentence: 'Buy 100 yuan of Tesla for me' or 'Buy one lot of Pop Mart for me,' and the trade can be completed. It is no longer complicated, and it will not be frightening.
I have always felt that everyone has their own way of changing the world. And if, because of products like Bobby and RockFlow, more people can turn the cognition from life and work into their own investment decisions, and increase penetration by 100 times—I think my life will have been worth it. This is truly my heartfelt sense of mission: to let everyone experience the joy of investing, and let them know that they too can participate in a company's growth and profit from it.
Actually, this principle is not hard to explain. Although there will certainly be people in the comments cursing, saying things like 'survivorship bias.' First, it is true that not everyone can make money investing; this must be acknowledged. Second, what is the key? I know so many hedge fund managers who missed Pop Mart, Laopu Gold, and Mixue Bingcheng. Why? Because ordinary people are the ones who will truly buy these products. These purchases are their lives. And many fund managers will not observe these things.
So fundamentally,Don't underestimate everything you see in life. For example, if you see an endless stream of customers at the entrance of a Yiming snack shop, you might think it's just an ordinary scene. But I'm telling you: this is your opportunity.Because what you see, Wall Street cannot see. You've long known that Bilibili's bullet comment culture is very popular; the young people around you use it every day. At the time, many primary market investors couldn't feel it. Many peopleby default think "investing" is a matter for professionals, requiring reading financial reports, analyzing volume-price relationships, and running models. As a result, they instead ignore the most valuable clue—your own life.
Actually, what is DAU? It's just more and more people around you using it. And what is growth? Isn't it just you constantly buying blind boxes? The reason I bought Pop Mart is because I saw more and more blind boxes piling up in the office, and everyone was asking every day how to buy Labubu. Mixue Bingcheng is the same. I never drink it, but my colleagues were saying "it's both tasty and cheap," so I went to subscribe to the IPO. In the end, I got an allotment, and it felt like the returns from the IPO subscription could let me drink Mixue Bingcheng for a lifetime.
👩🏻 Vakee
I remember that Sushiro case. Someone really liked eating at that restaurant, so he went and bought Japanese stocks of Sushiro's parent company. Because he observed—stores opening one after another, everyone eating there. So he bought. He was an ordinary person; he just made a decision based on his own life. That's why I always emphasize: opportunities in life are the best opportunities for ordinary people.
First, don't assume that investing must involve looking at volume-price indicators and studying financial reports.
Second, your observations and insights are not something everyone can possess.
Each generation has its own opportunities. If you can't even see the opportunities in your own life, and can't even convert the insights from your work into judgment, then you're even less likely to seize opportunities beyond your cognition.
At the end of 2023, I bought a lot of Robinhood stock at 8 dollars. If I'm building RockFlow yet don't even buy stock in the industry leader Robinhood, don't know when to buy, or how to buy, then as the founder of this AI trading platform, I'd be unqualified. Similarly, if you're an AI investor or entrepreneur, you must look at how much Nvidia is in your portfolio. Because primary market investing, secondary market investing, and entrepreneurship all share one common requirement: judgment.
What is judgment? It's your judgment about industry scale, direction of evolution, and changes in the competitive landscape. And this ability is best reflected in your positions in the secondary market.
👦🏻 Koji
Although opening a stock trading app is indeed not difficult, for many people, it is still a threshold that keeps them outside of "investing."
And Bobby's goal is to lower this threshold a bit further. So I think it really can enable many young people, for the first time, to truly and potentially enjoy: converting their own lives and insights into investment returns. This is also one of the positive values AI can bring us.
👩🏻 Vakee
Yes, I think so too. I really hope that because of Bobby's emergence, investing can "fly into the homes of ordinary people." Everyone can participate in the changes of this world through investing.We all know that the truly greatest returns are actually capital returns. We are consumers of many companies. For example, I subscribed to Duolingo, so of course I also bought its stock. In this way, I am both a consumer and a shareholder—I use its products, create revenue, drive up the stock price, and in the end I also get returns myself.
But the reality is,most users are only consumers. They create value for the company, but do not participate in capital gains., this part was profited by me because I hold its stock. You put in the effort, I make the money—this loop is actually incomplete.
But in fact, the ideal business loop should be: users are both consumers and shareholders. Because consumers are shareholders, they are more loyal; they will give you feedback, help you optimize the product, support you, and hope you do well.
We have already seen some companies start down this path, such as Tesla, which is a very typical example. Also, when Square went public, the investment bank invited many of its customers to subscribe to IPO shares. Because everyone holds stock, their cooperative relationships with each other also become more stable. This is a very positive loop. This includes Robinhood; many of its users are also its shareholders. In the future, if Xiaohongshu goes public, it should also let its users become shareholders.
Because once you understand this loop—the more you use it, the more the company's stock price rises, and the higher the return you get—of course you will support it, right?I even told Bilibili before that your premium membership should just include one share of Bilibili stock, given directly on RockFlow. This is the power of a virtuous cycle.
I think the most important thing is: to let ordinary people identify the trends around them, then buy companies they are familiar with, companies they have consumed from, and support products they like, become shareholders of these companies, and grow together with them. This is what a healthy commercial society looks like. The same goes for Didi drivers. They have contributed enormous value to Didi, but they do not have Didi stock, only wages from which commissions are deducted. This return system is unbalanced. But if buying stocks becomes simple enough, this loop can be established.
I think this is business equalization. And this is the change I truly want to drive through products.
👩🏻 Ronghui
I also want to add one final point: I think many people do not fail to see these signals; they actually do see them, but what they lack is 'a rational gatekeeper.' That is, a voice telling them, 'This can be done,' 'You can participate.' Just like you said earlier, a shift in awareness must be completed—I am not only a consumer, I can also be an investor. But in this process, people need rational confirmation. They will worry: 'Is this too professional?' 'Might I judge wrongly?'
👩🏻 Vakee
Right, so Bobby is not a voice order-placing feature; it is an AI Agent. Even if you just casually say, 'I want to buy 3,000 US dollars' worth of stock,' it will also have a complete chain of logic to tell you: Why can these stocks be considered? Why these 6? It will not just take instructions; it will help you explain and make judgments—using a way you can understand, turning ordinary cognition into executable investment suggestions.
For example, if you asked it last year, 'Labubu is very popular, what should I do?' it might tell you that you can buy Pop Mart. But if you ask now, it might say the price is too high and you should pay attention to risk. So Bobby is not just an order-placing tool; it has its own logical framework, risk judgment, and will also consider factors you did not think of, such as the current stock price trend, volatility, position level, and so on. These are actually places where ordinary users cannot judge independently, and these are all capabilities of Bobby AI.
Bobby has never promised to 'help you make money' or 'make money for you.' What it really does is—help you think one step further and manage risk. This also reflects our values. We never tell users: 'RockFlow can help you make money,' 'We can definitely make you money.' There is no such promise. Even I myself often say: if you copy-trade me, you may also lose money.
So I often say: you can start by trying it out with 100 dollars to get a feel for it. If you feel more confident in your investment ability, then increase your investment. We even specifically built a "position protection" feature—when the size of a single stock in your portfolio reaches double digits, such as 10%, the system reminds you: remember to pair it with a put for protection. We can't help you make money, but we will do our utmost to remind you, protect you, and assist you, so that every step you take is more steady. This is how our values are put into practice.
And actually, when it comes down to it, investing is a game of "staying at the table." As long as you are not eliminated, you have a chance to wait for the hand that belongs to you. So staying at the table is the most important thing, and risk management is always the top priority. Don't go all in unless you are very certain; even then, you must have protective measures to hedge risk.
👦🏻 Koji
Thank you very much to Vakee for joining us today on"Crossroads". Especially this last part—actually, since its founding, Crossroads has been saying one thing: "Seek out and bring together positive actors in the AI era." And one manifestation of "positive action" is—after listening to today's episode, are you willing to turn your understanding of the AI era into investment action?
And Bobby is exactly the tool that lowers the threshold for this action. I recommend everyone go try it. Although Bobby does not promise to help everyone make money, we also hope that with its companionship and assistance, everyone can really earn a little investment return.
👩🏻 Vakee
Alright, thank you everyone. I'm very happy to share with you today, and I look forward to more people starting their investment journey because of Bobby, and then being able to feel the joy of investing. Thank you.
Original link:The first investment trading Agent Bobby launches, first interview with Vakee