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

This article is republished with permission from AI NOW! (ID: ainowainow). Produced by: AI NOW!

Before building Bobby, Vakee had been trading stocks for over 20 years.

She started at age 9 (three years later than Warren Buffett) and could execute trades on pure intuition. After graduating from Imperial College London, she joined a quant fund, using machine learning to trade futures and derivatives. In her twenties, she made her first 10 million by shorting a U.S. stock.

The numbers in her account ballooned like a video game score. For Vakee, making money was too easy — almost boring. She had little interest in effortless pleasures: buying designer bags, traveling, hoarding luxury goods. She typically shopped on Pinduoduo, never flew business class unless it actually got her there faster.

When trading became as effortless as breathing, she needed to find a more thrilling game.

After several years doing product and R&D at a major tech company, followed by five years in venture capital, Vakee officially launched her own startup. Her first product was RockFlow (Chinese name: 奇运证券), an AI-powered U.S. stock brokerage for global users.

RockFlow used AI to simplify investing, rolling out modules like daily trade recommendations, auto-copy trading, and streamlined options — transforming complex financial platforms into something as engaging as a game.

But this product, launched in 2021, was still fundamentally a "smarter tool." It couldn't understand why a user wanted to trade at a given moment, let alone place orders for them directly.

Now, Vakee is about to launch her next project: an AI Agent named Bobby.

Bobby's goal is to completely replace the app. Ordinary users simply express their ideas in natural language, and Bobby handles the full pipeline: intent decomposition → strategy generation → order execution — closing the trading loop.

RockFlow app order interface vs. Bobby natural language order page

For example: when you're browsing Pop Mart's page, Bobby might pop up: "You've spent 2,000 yuan on Pop Mart this month. Labubu is trending on TikTok. The Labubu x Vans collab is trading at a 1,284% premium on the secondary market. Want to increase your position in the designer toy sector?"

This represents a generational leap beyond chatbots that merely answer questions (like BloombergGPT or the in-development Morgan Stanley GPT) or smart tools requiring manual operation (like the original RockFlow).

In Vakee's view, Agents aren't upgraded apps — they're the next-generation user interface.

She even asserts: "All apps will disappear in the future. They'll all be replaced by Agents."

More radically: when you tell Bobby "I hate Donald Trump," Bobby will translate that sentiment into actionable trades — perhaps avoiding or shorting Trump-related stocks (like DJT), while simultaneously recommending long positions in clean energy based on his pro-fossil-fuel policy stance. It will ask about your investment expectations and help select assets matching your risk appetite and return targets.

A person's emotions and values are becoming trading instructions.

This is precisely the starting point for why Vakee created Bobby.

Vakee's love for "trading" is as natural as others' love for food or travel. For most people, investing is complex and intimidating. For her, every morning opens onto a world brimming with trading opportunities — available anytime, anywhere, effortlessly seized.

Take this past Spring Festival, when DeepSeek exploded and relatives back home started chatting with AI. The RockFlow team quickly applied for DeepSeek-related services from cloud providers, while she immediately bought Alibaba — because cloud computing would inevitably benefit from the compute expansion demand driven by open-source models.

Another time, traveling in Japan, she noticed Sagami's ultra-thin condoms were sold out at every convenience store shelf. She immediately researched the company, discovering its polyurethane ultra-thin exploration predated and outpaced Okamoto's, and that it topped cross-border e-commerce platforms in China. That stock ultimately returned 4x that year.

Trading is a lifestyle. Vakee believes this unequivocally.

She doesn't love indulgence — she shops on Pindouduo daily. Instead, she's nearly fanatical about anything requiring training, challenge, and extreme mobilization. She loves the rush of short-term projects: passing the bar exam in 9 days, scoring highest in her class for the prestigious U.S. ACE personal trainer certification.

Then she turned that passion toward AI. For the past decade, she's been either building AI products, investing in AI, or founding AI startups.

Now, Vakee has finally found a worthy new boss: Bobby.

The RockFlow team skews introverted — team bonding means sitting in silence, gazing at the sky

When we met Vakee, the tariff war had just begun. In the café, panicked market sentiment crackled through everyone's phone screens.

"What can Bobby do right now?" I asked.

"Bobby, find companies least affected by tariffs with potential for earnings surprises over the next three months," Vakee said.

Seconds later, Bobby responded:

"Consider tariff-exempt sectors such as AI infrastructure, data centers, and semiconductor manufacturing equipment — strategic competitiveness areas likely to receive partial tariff exemptions or delayed implementation. U.S. domestic production leaders may also outperform, as local manufacturing avoids import tariffs."

"Additionally, consider allocating to demand-inelastic industries and defensive sectors. Refer to the Tariff-Resistant Stock List."

And proactively asked: "Shall I auto-place orders if any of these stocks drop over 10% within a week?"

In the AI era, some use large models for videos and PowerPoints, some replace dating with DeepSeek chats, others want to raise an immortal AI cat.

But Vakee believes future trading will be AI versus AI — so everyone needs a Bobby.

---

AI NOW!: Many financial products have integrated large models today — Bloomberg's BloombergGPT, Morgan Stanley's developing AI assistant GPT Copilot. What's the difference between Bobby and them? A chatbot that can place orders?

Vakee: If Bobby were just a chat tool that can trade, I wouldn't have built it. Have you ever asked those financial chatbots "what should I buy now"?

AI NOW!: They analyze the market, give some advice, and inevitably end with "investment carries risks."

Vakee: Right, because they're answering, not deciding. Bobby is different. Say "I hate Donald Trump" — it won't give you a summary of "Trump's policy impact on markets." It will directly ask: "Shall I filter out Republican-linked stocks? Detected potential upside in clean energy — these are specific targets. Shall I adjust your position?"

AI NOW!: So this is advanced semantic understanding?

Vakee: No, completely different logic. Chatbots are: you ask, I answer. Bobby is: you haven't asked, but I'm already calculating. If it notices you've been reading semiconductor news but hold no chip stocks, it proactively asks: "Shall I monitor TSMC earnings? If they beat, consider buying?" Of course, you can also teach Bobby your own analytical logic.

AI NOW!: Sounds like Bobby solves the critical gap between a person's trading impulse and execution — is this hard for typical users?

Vakee: For many, yes. Take this recent U.S. market drop — before the tariffs, everyone knew a major drop was highly probable. But judging the specific timing, and what to sell and how much, requires diligent research and tracking details to act correctly. Most people have ideas but don't execute — maybe lazy, maybe unsure how to operate specifically.

Even the simplest strategy — buy low, sell high — many can't execute. At 30% gains, most won't sell, hoping for 50%. Missing profit-taking, then panicking and selling when it crashes.

AI won't, because AI has no human nature.

AI NOW!: Can you not make money in stocks with human nature?

Vakee: Human nature has too much greed, anger, delusion. Going against human nature is how you make money in stocks — so AI will definitely make money better.

AI NOW!: What if I said: Bobby, can you adjust my portfolio to zero risk, 100% return?

Vakee: See, that's greed, hahaha. Bobby will honestly say "impossible," then translate your get-rich-quick fantasy into an executable plan for 20% annualized returns.

AI NOW!: But won't trust costs be too high for a trading Agent? A PowerPoint Agent making mistakes can be fixed — worst case, it's wrong. Won't users worry about Agents losing them money?

Vakee: We set safety words. You can say "Bobby, monitor only, no trading," "call me if NVIDIA drops to 98," or "no single trade over $10,000." Most people actually feel more reassured after trying it — same reason, AI has no human weaknesses.

AI NOW!: Your first startup was RockFlow, a Gen Z brokerage platform. On that foundation — was Bobby absolutely necessary, or is it because everyone's jumping into AI Agents this year?

Vakee: My vision for RockFlow from day one was "make investing simpler." For two years, we tried many approaches — making the app extremely simple. But I realized it wasn't enough.

By 2025, I believe new-generation users, especially AI natives, can absolutely trade through natural language. You just speak, and Bobby rapidly realizes intent decomposition, strategy generation, through to order execution.

RockFlow's early NFT avatar design sketches

AI NOW!: Many brokerages now integrate general-purpose large models for investment advice. Why didn't you first add a chat box to RockFlow — say, plug in DeepSeek for stock picking? That seems like the most natural upgrade path.

Vakee: General models do provide standardized analysis, but this is far from sufficient for real investment trading. When we need precise research based on each user's personalized needs and trading preferences, with real-time requirements guiding immediate execution, general models are powerless. They can only batch-pass searched information to users, unable to complete a true decision loop.

Bobby uses a workflow + LLM/Agent model, maximizing AI creativity while controlling cost and risk.

The key is: all information in our workflow — whether user data, market price/volume data, or capital market sentiment data — undergoes secondary processing through our professional know-how. Only then can we generate responses that are both sound and truly user-understanding, ultimately completing trade execution.

AI NOW!: Many general Agents are also integrating vertical domain knowledge bases — Coze Space claims to connect to certain expert systems. Why do you insist that vertical Agents like Bobby are the future, rather than waiting for general Agents to grow stronger?

Vakee: This comes down to a fundamental question: there are two types of needs in the world.

First: life-or-death needs. Like financial trading, medical diagnosis — most people can't score 70 on these, and scoring below 70 equals zero, because the consequences are severe. These tasks have extremely high professional barriers, extremely low fault tolerance. Yet simultaneously, there's a clear methodical path to scoring 70.

Second: nice-to-have needs. Like making PowerPoints, writing travel guides — ordinary people scoring below 70 is fine, it doesn't equal zero, just less than perfect.

General Agents suit the second type. But life-or-death needs like financial trading must be handled by vertical Agents. Several reasons:

• Data dimensions differ: we process millisecond-level market data, real-time user positions, and other professional information
• Accountability levels differ: one wrong investment recommendation could cause major user losses
• Decision mechanisms differ: not giving "maybe possibly" advice, but making executable judgments

AI NOW!: Like you wouldn't let a general practitioner do heart surgery?

Vakee: Right, you wouldn't dare. Bobby was born a "finance major" — every judgment rests on professional trader-level training.

AI NOW!: Many think workflow can only handle narrow tasks like simple queries. How do you make Bobby truly "understand finance"?

Vakee: Workflow itself is just a tool — what matters is how you use it.

Industry knowledge is the foundation. The real breakthrough: we have workflow dynamically generate trading information most relevant to the user's current moment — including risk appetite, real-time holdings, trading intent, and market sentiment — combined with our team's accumulated financial know-how, instantly composing through thousands of dynamic nodes. This isn't simple information retrieval; it's like a professional trader, parsing and responding to market signals in real time. In RockFlow's validated investment scenarios, this system's execution efficiency exceeds general large models by orders of magnitude.

Our ideal model: build a data-driven, auto-evolving world model, enabling financial decisions to continuously learn and adapt in dynamically changing markets, achieving true intelligence and efficiency.

AI NOW!: What about the "understand user" side? How does Bobby know who you are and what you want?

Vakee: Bobby connects to RockFlow's counter trading system, real-time market data, and user data — like a hedge fund manager on 24-hour standby.

For example, when a user says "I just got laid off, want stable investing," Bobby automatically lowers risk appetite, recommends treasury bonds + high-dividend stocks, and sets dynamic stop-losses. It might proactively ask during market volatility: "Detected potential Federal Reserve rate hike — adjust your bond position?"

AI NOW!: If vertical financial Agents are the future, why haven't we seen Robinhood, Futu Holdings Limited, and others moving this direction?

Vakee: The AI-native investment trading platform experience requires a fundamentally different technical architecture from mobile internet brokerages from day one. The more successful incumbent peers are, the harder it is to abandon all existing business infrastructure and user experience for AI-era self-revolution.

The previous generation mobile brokerages' mission was providing excellent mobile client experiences — a massive improvement over PC-era incumbents like Interactive Brokers. They did this well, satisfying 70s and 80s generation needs. But in the AI era, GenZ and younger users have new investment trading experience demands.

RockFlow's difference: from day one, we explicitly targeted new-generation investors with an AI-native wealth management platform, building our own AI infrastructure and AI-training-suitable counter trading system. Only we are doing this worldwide.

AI NOW!: Wait — what does building your own counter trading system mean?

Vakee: For a brokerage platform, the counter trading system is like Douyin's recommendation algorithm. Using someone else's system is a black box — not every module opens for model training. So we had to build our own. Only then can we obtain structured, continuous user behavior data from the ground up, understand users' true decision paths, and continuously feedback-optimize.

When we first designed RockFlow's counter trading system, we already considered how each module would do machine learning in the future. This data isn't static assets — it's all raw material for training each person's personalized Bobby.

AI NOW!: Sounds like you were preparing for Agents from day one.

Vakee: Checked our first meeting notes about Bobby — September 2023. We designed RockFlow's entire system from the start toward AI-native goals, including trading system, data, and product architecture. But the thinking clarified gradually; we also took detours in Agent architecture design — all valuable experience.

Every generation of user-facing endpoint products has its historical mission. In the AI era, your product mission certainly isn't just slightly richer features, simpler design interactions, adding or removing a button.

AI NOW!: What's the AI-era product mission?

Vakee: I believe it's the first real possibility to understand users and proactively serve them. Mobile internet's good products let users "operate more efficiently"; the AI era lets users not operate — directly receive service.

So on the app, it might step-by-step guide you "how to buy options," "how to find suitable options." But today AI directly recognizes your intent — "I want 20% investment returns," "I want to survive this market crash." This is a fundamentally revolutionary paradigm shift.

So I believe Agent isn't an upgraded app — it is the next-generation user interface.

AI NOW!: Specifically for Bobby, what's its mission?

Vakee: My original intention is believing investing is highly personalized — it's completely a comprehensive expression of one's values and worldview.

As we discussed, many people have ideas, they have various cognitions about daily events, but they're trapped by operational details preventing correct and timely action. People often say trading is cognition monetization — but for most, the biggest difficulty is from cognition to trade, they don't know how to operate. Bobby exists to solve this.

AI NOW!: Whether I hate or like Donald Trump, I can trade it.

Vakee: All tradeable, all profitable.

AI NOW!: From operating a tool, to being served by an agent.

Vakee: Right. In this sense, all future apps will disappear, replaced by Agents.

AI NOW!: I believe this future too. But have you worried you're too early, becoming an industry pioneer (who dies)?

Vakee: I don't think that way. For 10 years, I've either been investing in AI or founding AI companies. All my experiences determined that today I must do this one thing: Bobby. The brave enjoy the world first.

---

• Biggest AI shock in 2025

Vakee: When DeepSeek made its deep reasoning visible to everyone.

• In the past year, what in AI did you initially dismiss but completely changed your mind?

Vakee: Text-to-image. Accurately generating images — ensuring consistent character poses, meaningful text — was extremely hard initially. Thought image generation was very far from commercialization. Then new tech and products from Diffusion Transformer onward were incredibly powerful, far exceeding expectations. Now image generation is fully usable across production scenarios.

• If you could personally shut down one AI product or trend — something you think is completely pseudo-demand or wrong direction — which would you choose?

Vakee: Building general features on top of OpenAI-type companies, or products not deeply embedded in business scenario closed loops — these are very difficult. OpenAI's new models easily disrupt them: when GPT-4o's one-click Ghibli-style generation came out, many startups died, having failed to think through how to build business moats.

So for me, what matters isn't "am I using AI," but "what problem am I solving" — whether you can abstract needs in vertical scenarios while considering long-term commercial value and business barriers. The industry currently lacks more excellent AI product managers.

So returning to Bobby: building Bobby isn't to show off AI tech, but to make investing simpler, to create value. If one day I find Bobby isn't achieving this, it can absolutely be killed too — no need for attachment.

• At AI's current development stage, what's undervalued? What's overvalued?

Vakee: Compute demand is undervalued; AGI arrival is overvalued. Now everyone thinks stacking more GPUs, training larger parameter models will approach AGI. But the real bottleneck is actually the application layer — I believe the coming years will be vertical Agents' explosive period, and this is a very long-term process. Every vertical domain needs customized compute optimization. Like after EVs proliferated, charging stations became the real bottleneck.

• What are you most looking forward to in 2025?

Vakee: The "Cambrian explosion" in vertical application domains. In complex scenarios like finance, healthcare, education, travel, and supply chain, Agents that truly reconstruct user experience will emerge — not just simple chatbots.

For example, travel Agents that automatically complete personalized trip planning, negotiation, and payment; or education Agents that customize learning paths based on learning ability and preferences. These don't require waiting for AGI — existing technology plus vertical data can achieve them.

• Finally, recommend three favorite books!

Vakee: Richard Feynman's The Pleasure of Finding Things Out, Dzongsar Jamyang Khyentse Rinpoche's What Makes You Not a Buddhist, and Ezra Vogel's Deng Xiaoping and the Transformation of China.

---

Intelligence Surging: BlueRun Ventures' AI Portfolio Companies Launch New Products Again! | BlueRun Family Headlines

Planting Spring in the Wind | Quarterly

Making Cars Intelligent Agents: Li Auto Releases Next-Generation Autonomous Driving Architecture MindVLA | BlueRun Family Headlines

Embodied Intelligence Enters Reality | Path to AGI