Interview with Bill, Head of Bitget AI: In the era of AI trading, how far are we from "making money while lying down"?

Author: Frank, PANews

A "crayfish" has stirred up the entire tech world. The sudden emergence of OpenClaw has everyone excited. On an ordinary personal computer, AI can be given operational permissions to help you receive emails, write code, and even manage trading accounts. The internet is flooded with case studies describing it in fantastical terms: "You could quit your job!" But after actually installing it, most people find that things aren't quite like that.

In the crypto trading space, this shift from frenzy to calm is particularly pronounced. Over the past two years, almost every exchange has launched its own "AI Agent," but most have remained at the stage of chat assistance—you ask it a question, it generates a lengthy analysis, and that's about it. The emergence of OpenClaw is like opening Pandora's box, showing everyone the possibility of AI "doing things" rather than just "speaking."

But this has brought new challenges. As a leading figure in exploring the forefront of AI trading, Dr. Bill, head of Bitget AI, has a deep understanding of this, and PANews conducted an in-depth interview with him. Before joining Bitget, Bill held senior positions in several leading internet and technology companies, leading the large-scale deployment of several core algorithms and AI platforms, and publishing dozens of papers at top international conferences and obtaining dozens of patents.

Today, he is fully responsible for Bitget's AI strategic planning and intelligent trading technology research and development, and is committed to promoting the deep integration of AI and crypto asset trading scenarios. Faced with the current agent craze, this leading expert's assessment is remarkably calm: "Most ordinary people are not used to being managers. Suddenly giving someone 10 AI subordinates—how to direct, assign tasks, and evaluate them—is itself an art. "

The initial enthusiasm will eventually fade, but the ability has been recognized. The real question then becomes: who can package this ability into a product that ordinary people can use?

In a conversation with Bill, PANews attempted to analyze the real path of AI trading from concept to reality from the perspective of a product designer. According to Bill, Bitget's rapid launch of two AI products, Agent Hub and GetClaw, was not simply a case of "seeing others do it and doing it too," but rather a natural spillover effect from an internal product. "In short, it was a perfect storm of favorable timing, location, and people."

The right timing was that OpenClaw ignited market awareness; the right location was that we had accumulated deep experience through continuous iterations of our AI assistant GetAgent, launched last year, and had sufficient internal technical expertise and experimentation; and the right people were that the team had validated the product's value internally and was now opening it up to the outside world.

Bitget's AI Product Overview: A Three-Tier Architecture from GetAgent to GetClaw

To understand Bitget's strategy in AI trading, it's essential to first clarify the relationship between its three products. While the names GetAgent, Agent Hub, and GetClaw may seem confusing to outsiders, Bill explains that it represents a clear evolutionary path.

In June 2025, Bitget launched GetAgent within its app, an AI trading assistant in the form of a chatbot. According to Bill, GetAgent has undergone multiple iterations: from initial chat responses, it gradually added one-click order placement and news aggregation, and then expanded to cover all categories of trading, including US stocks, gold, and silver. "Each iteration is driven by user needs, expanding more and more each time." But no matter how it expands, GetAgent's essence remains "chat-driven"; it can answer questions and give advice, but it cannot help users independently execute complex trading tasks.

The turning point came after the release of OpenClaw. According to Bill, after the release of OpenClaw, Bitget quickly built its own version internally. "The feedback after internal use was very good, which naturally led to the idea: could GetAgent also undergo a major upgrade?" Following this line of thought, Bitget packaged its internally refined MCP capabilities and opened them up to the outside world, officially releasing Agent Hub on February 13th of this year.

Agent Hub is geared towards professional players who have "relatively strong hands-on skills".

It provides four layers of capability interfaces, from shallow to deep:

APIs are atomic interface calls, which have the highest barrier to entry, requiring programming and key management.

MCP acts as a "universal interface," allowing external AI applications to directly read Bitget's data and perform operations.

The CLI is geared towards developers and supports direct calls to all APIs via the terminal command line.

Skills are the core of this upgrade, essentially encapsulated "business modules." Through Skills, the originally rigid API code has been transformed into skills that AI can directly invoke (such as querying rates, analyzing candlestick charts, monitoring the market, and placing orders). AI has thus achieved a leap from "intent understanding" to "action execution."

Bill used a USB flash drive as a very intuitive analogy: "A USB flash drive itself has the storage capabilities of storing, reading, and writing, but to make it work, it needs a USB interface to connect to the device, which is equivalent to an MCP. But having an interface alone is not enough; it also needs the cooperation of storage and various protocols to complete the full interaction. This whole combination constitutes a Skill."

However, Agent Hub still has barriers for ordinary users.

So on March 14th, Bitget launched GetClaw, an AI trading assistant based on Telegram. It's ready to use out of the box, requiring no installation. Users access it via a link, log in to their account, and start using it. The platform bears the cost of calling the large model, so users are completely unaware of it. Bill summarized it in one sentence: " For ordinary users, I recommend using GetClaw, a pre-assembled tool that you can start using immediately; for professional players, I recommend using Agent Hub, choosing the appropriate skills, and building your own castle like assembling Lego bricks. "

These three products form a clear progression: GetAgent refined the underlying MCP capabilities, made them available in Agent Hub, and then embedded these capabilities into GetClaw, lowering the barrier to entry to the bare minimum. From chatbots to developer tools to one-click products, Bitget's AI product line covers the entire user spectrum, from tech enthusiasts to beginners.

"Monitoring the market with just one sentence": What has AI trading truly changed?

Product architecture is just the skeleton; what truly excites users is the transformative experience that AI brings in specific scenarios. In conversations with Bill, a recurring keyword was "threshold."

Traditional trading processes are lengthy chains: information gathering, analysis and decision-making, order execution, market monitoring, and post-trade analysis, each step relying on manual operation. If users want to implement conditional trading or quantitative strategies, they either have to write their own programs to call APIs or configure a host of complex parameters on the platform.

In Bill's view, this is precisely the most valuable entry point for AI: "These functions can be achieved without skills or getclaws; you just need to write a program. But the problem is that writing programs is simple for programmers, but the threshold is too high for ordinary users. What we are doing today is enabling users to achieve the same effect by simply saying a sentence."

He gave a specific example: a user says, "When Bitcoin drops 3% in one minute, help me add 50% to my position," and the system automatically converts this into a scheduled task. This task actually needs to complete three things:

1. Real-time monitoring of Bitcoin prices

2. Calculate price difference per minute

3. Once the conditions are met, immediately execute the additional position operation.

This kind of logic, which used to be only possible for programmers, can now be accomplished by anyone with just a few words.

Within 40 hours of its launch, GetClaw saw its market monitoring alerts become its most popular use case. This is not surprising. On traditional platforms, configuring market monitoring alerts required users to understand various indicator parameters, and "it could take a long time to configure and still not be successful." Now, even for multi-indicator complex monitoring logic such as MACD and CCI, users can describe their needs in natural language, and the system can handle the task for them.

But Bill believes that the real revolution of AI-powered market monitoring lies not only in its ability to "do it," but also in its ability to "optimize it." "On traditional platforms, if it doesn't work, you give up. But now you can tell it, 'It's wrong, let's reflect on how to fix it,' and keep working on it until you're satisfied." This interactive method, which allows for continuous iteration, is a huge satisfaction for the vast long-tail user group.

In traditional stock markets, quantitative trading is becoming increasingly prevalent, even exceeding 70% in the relatively mature US market. Ordinary retail investors entering the market face institutional competitors operating at a microsecond-level pace, with virtually no chance of success. Bill summarizes the significance of AI trading as a form of "equal access": " Bitget's vision in the field of AI is to enable 100 million users to rival Wall Street , in other words, to achieve the operational logic and execution capabilities of top traders. In the past, it was something we could think of but couldn't do; today, if we can think of it, we can do it."

Four Locks of Trust: Security Boundaries When AI Operates with Real Money

As AI moves from "giving advice" to "executing for you," the biggest challenge isn't the sheer power of its features, but rather trust. Bill believes this point cannot be overstated: "The biggest concern for ordinary users is 'Is it safe to use?' This level of trust must be firmly established. Once a security issue arises once or twice, nobody will use it anymore."

To address this core concern, Bitget designed a four-layer isolation system.

• The first layer is identity isolation, which accurately identifies the user's identity in each conversation.

• The second layer is memory isolation, where conversation memories between different users are completely isolated and obfuscated, ensuring that personal privacy is not leaked.

• The third layer is access control, which determines what data and tools can be accessed, controlled by roles.

• The fourth layer separates credentials from funds. API keys are only used for triggering transactions, and transactions are executed within a sub-account sandbox.

The sub-account sandbox mechanism is a very practical design. Bill gave an example: "For instance, if the main account has $1,000, the user can transfer only $50 to the sub-account for the AI ​​to operate, which makes the risk much more controllable." This means that even if the AI ​​makes a judgment error, the risk exposure is strictly controlled within the range preset by the user.

This security-first approach is also reflected in Bitget's attitude towards the Skills Marketplace. Currently, all Skills are developed and maintained by Bitget and are not open to third parties. Bill's explanation is straightforward: "If we open the Skills Marketplace and allow more people to participate in its development, security issues will inevitably arise. For example, if a hacker says, 'I'll put one in for you too,' and users lose money after using it, that's not appropriate. We call it 'better to have none than to lose all your money playing around. ' After all, in the asset market, it's better to survive long enough to earn quickly."

The cautionary tale of OpenClaw justifies this approach. While its near-unrestricted operation on personal computers was exciting, it also spawned an absurd new industry, with "cleanly uninstalling Lobster" itself becoming a lucrative business.

At the large model invocation level, Bitget initially chose to have the platform bear the costs rather than allowing users to configure their own tokens. This was partly for security reasons and partly for technical reasons. "Our Skills and MCP have been deeply adapted and optimized with various built-in large models. If users switch to other models at will, the effect will be greatly reduced." Currently, the platform provides each user with a free invocation quota of $10 per day, and will adjust the pricing model based on market feedback in the future.

80% of the work can be done, but 20% of the decisions still depend on people.

When discussing the realistic limitations of AI trading capabilities, Bill frankly admitted that the reality is not optimistic: "There are people online who give AI $100 to make $1,000, but they find that this kind of crude operation has a very high probability of losing everything."

AI trading capabilities cannot guarantee profits for users today. Bill uses the "Pareto Principle" to summarize the current reality: in a complete trading process (which may involve 100 tasks), AI can efficiently complete 80 of the more complex tasks, such as information organization, real-time monitoring, conditional execution, and data review. However, AI cannot yet handle the 20 core decisions that truly determine profit or loss.

Last year, Bitget held a casual AI trading competition to test the limits of AI's capabilities. The results provided a vivid illustration: many AI strategies ultimately ended in losses. The reason is not complicated: AI lacks emotions, which sounds like an advantage, but it also means it cannot respond to extreme black swan events such as a sudden war. Bill mentioned that when AI was heavily involved in trading in the US stock market in the past, there were also abnormal phenomena such as sharp drops and surges.

“Today, it’s more of an advanced assistance function, like the transition from Level 1 to Level 5 autonomous driving,” Bill used this analogy to describe the current stage of AI trading development. Looking at the trend, AI is indeed overcoming the remaining challenges one by one, but when it comes to long-term creativity and empathetic judgment in extreme situations, machines still have significant bottlenecks.

However, Bill also offered a relatively optimistic assessment: "The technological loop surrounding fully automated trading may be largely realized next year, but this does not mean it can guarantee continued profitability." In other words, there is still a considerable gap between "being able to run" and "being able to make money."

From trading tools to an "AI account operating system," Bitget's ultimate vision.

Since AI cannot completely replace human traders in the short term, where does Bitget's AI strategy end? Bill gave a three-dimensional answer.

The first dimension is "panoramic trading," which echoes Bitget's previously proposed UEX (Universal Exchange) strategy. Beyond cryptocurrencies, with the advancement of asset tokenization, traditional financial categories such as gold, silver, and US stocks are being integrated. Bitget aims to use AI to help users complete all categories of trading operations on a single platform, "giving users the comprehensive trading capabilities of Wall Street traders."

The second dimension is global ecosystem expansion. By leveraging Bitget Wallet's capabilities, AI is introduced into Web3 payments and global business scenarios, lowering the operational barriers to cross-border transactions and payments.

The third dimension, and the one Bill described most vividly, is building a "long-term account operating system" based on Bitget. The core of this concept is to establish a "high-trust fund execution layer," in which multiple agents will collaborate to help users with various tasks. The foundation supporting all of this is a cross-platform, cross-scenario "long-term memory system."

In Bill's description, this memory system analyzes and integrates a user's past trading habits, historical operations, and even their subtle behaviors within the app, forming a deep personal profile. "It ensures that the user's trading logic remains consistent across different platforms and scenarios, rather than providing a fragmented experience." This ability to continuously learn and adapt is the fundamental difference between this system and disposable tools.

He used a very common analogy to explain this gradual trust process: "Just like when you first buy a housekeeping robot, you only use it to sweep the floor. After using it for a long time and trusting it, you are willing to let it take on more tasks." AI needs to prove its reliability in small tasks first, and then gradually gain greater authority and trust. The ultimate goal is to "accompany you as you grow and help your assets appreciate."

From GetAgent to Agent Hub and then to GetClaw, Bitget's AI products have made a leap from chatbots to task execution layers in less than a year. The intensive deployment of AI by major exchanges also indicates that AI trading is no longer an optional field, but a fundamental competitive capability for the future.

However, in reality, AI excels at replacing the "physical labor" rather than the "mental labor" in trading. While 80% of the complex tasks can be handled by machines, the crucial 20% of judgment that determines profit or loss will likely still require human intervention. Technology can lower the barrier to entry for trading, but it cannot completely eliminate trading risks.

AI has given everyone access to Wall Street's toolbox, but that toolbox contains both opportunity and awe.