This episode of the podcast focuses on Bitget's AI trading product layout. Dr. Bill, the head of Bitget AI, reflected on his transition from traditional AI research and industry experience to the crypto sector, and systematically introduced Bitget's iterative path for AI trading products over the past year: from initially helping users capture market information and organizing news and signals, to generating risk profiles and personalized suggestions based on user historical behavior, and finally attempting to lower the threshold for using AI trading through Agent Hub, Telegram formats, and interactive methods similar to Claude Code.

The interview also discussed the boundaries of AI in trading: it has significantly improved the information processing and decision-making efficiency of ordinary users, but still struggles to completely replace top traders; in the future, the competitive focus will not only be on model capabilities but also on security systems, cost control, product fluidity, long-term memory systems, and continuous learning about users' real trading habits. Both parties finally explored whether AI trading could lead to a "winner takes all" scenario, and whether strategies would quickly become ineffective. The conclusion is that the market remains highly complex, and human nature along with black swan factors still make it impossible for any single system to dominate trading completely.

Dr. Bill's AI Background and Entry into the Crypto Industry

Cat Brother: Welcome everyone to this episode of "Wu Says No to Crypto Podcast." Today's interviewee is Dr. Bill, the head of Bitget AI. Please introduce yourself and how you entered the crypto industry? I'd also like to hear about your experience in AI, as I hear everyone calls you Dr. Bill. Did you come from an AI background?

Dr. Bill: I graduated with a PhD in 2009, with both my undergraduate and master's studies focused on AI. During my studies, I also visited several companies and research institutes for exchanges and attended many international conferences.

After graduation, I worked for four years in an overseas research institute focused on artificial intelligence R&D. Later, I joined a major domestic company and worked there for four years in search recommendation and natural language processing, eventually leading the natural language processing department. After that, I went to an overseas e-commerce firm for four years responsible for overall AI R&D, and then spent three years at another large company managing global marketing algorithm R&D. In total, I have about sixteen years of experience.

At the beginning of last year, I was contacted by a headhunter who mentioned an opportunity at Bitget. Although I had not worked in the crypto sector before, I have always been quite interested in finance and have traded US stocks and Hong Kong stocks for many years, so I decided to give it a try.

At that time, I wasn’t very familiar with Web3; I only had some understanding and had not actually done relevant work, so I was a bit nervous before the interview. However, I passed the interview quickly and received an offer. My position is the head of AI at Bitget, and it has been over a year now. Overall, this experience has been quite exciting for me. There are new challenges and projects every day, which, although highly stressful, provide a great sense of accomplishment.

For me, the biggest change has been the cognitive impact. I originally only knew about Web3 by hearsay and hadn’t really participated deeply, so once I entered, I was essentially learning while doing projects, which has been very fulfilling.

Is the Combination of AI and Trading Just a Gimmick or Has It Entered a Practical Stage?

Cat Brother: Bitget is a platform primarily focused on trading products. What is your view on the "AI + trading" phenomenon? Is it really entering a feasible stage, or is it more of a market hype? Because now, it's not just the crypto industry; almost every industry is embracing AI. Returning to this topic, do you think its practical application is primary, or is there still an element of riding the trend?

Dr. Bill: I believe that for Bitget, this is no longer a gimmick but a necessity. Because for the first seven years, Bitget did not have a dedicated AI team, and there were very few algorithm applications; it was only in the last two years that we began systematic investment, with the core reason being that AI has matured enough to truly enter trading scenarios, whether for cost reduction and efficiency, or for enhancing income and trading efficiency, it already has practical value.

Trading itself is very complex; different users have varying levels of cognition, risk preferences, strategies, and methods of operation, so the key is not whether "AI should be integrated", but rather at which layer AI should intervene in the trading chain.

If it’s about full automation, like fully autonomous driving, I think that is still not achievable at this stage; however, assistance that segments and layers is already very feasible. In fact, regardless of whether Bitget does it, other companies are also doing it and have already reaped several benefits.

For instance, some traders mainly focus on short-term trends and quantitative signals, and previously they needed to monitor many screens and data; now AI is very suitable for integration and assisting in judgment. There are also those who base their decisions on news, financial reports, and social media; much of this work is actually about information collection and organization, which AI can visibly enhance in terms of efficiency.

Furthermore, users may also want AI not just to find information for them but to provide more specific strategic suggestions, such as position sizing, direction, leverage, and even preparing the trading buttons. At a higher level, it could even approach an asset management model.

So our judgment is that AI cannot completely replace the top professional traders, but for ordinary users, achieving 95% of the work equivalent is practically in a usable stage today.

Evolution of Bitget's AI Products: From Information Organization to Trading Assistance

Cat Brother: Are you suggesting that the first layer is quite mature, such as helping users understand project backgrounds, organizing information, and assisting in judgments? So, does Bitget’s current AI product lean more towards early decision support or has it already moved towards actual execution?

Dr. Bill: This goes back to last year. A month after I joined, we initiated the Agent direction. At that time, Agent was still a relatively new concept, and everyone was exploring. We first made a small attempt called "Meme Catcher" because Meme coins were particularly popular and the market signals were fast and mixed, making it difficult for users to seize trading opportunities in a timely manner.

This product was developed over two months and performed reasonably well, but its capability was relatively single, mainly capturing Meme-related signals. Later, we upgraded it to GetAgent, with the initial goal of solving the first layer of needs, which is information collection and organization. Because this part is essentially labor-intensive, optimizing the processes and models can significantly enhance efficiency.

So initially we focused on information-related capabilities, including customizing important news sources in the crypto space, then providing that high-quality information for model analysis, rather than letting the model search across the entire web alone. This approach led to substantial improvements in the accuracy of information collection and analysis, leading to higher user satisfaction.

However, users later began to express further needs; they not only wanted to view information but also sought decision-making suggestions, such as whether to go long or short, how much to buy, and what risk-level strategies to use. Thus, we started to analyze user historical trading records to create profiles to assess their risk preferences and trading habits, then provide more personalized suggestions.

Because the information layer can be relatively universal, the differences become significant at the trading level. Different users might have completely different answers to the same problem. So later, GetAgent gradually moved towards personalized matching, refining many details in the process.

We even managed to reach the execution layer. For example, users can directly say "Help me buy 10U of Bitcoin," and the system will quickly prepare the trading button; after user confirmation, the order can be placed. Of course, the premise is that the command must be clear enough and not too vague.

After this function was launched, there were indeed users utilizing it, and trading volumes increased. But subsequently, we found that if we pushed further towards "directly helping users place orders," users were likely to misunderstand and think that this product could guarantee them profits. Once losses occurred, there could be issues arising between expectations and reality.

Therefore, we adjusted our direction and did not continue to focus on optimizing automatic ordering; instead, we shifted our focus back to information collection, aggregation analysis, and personalized supply to solidify those capabilities.

This year at the beginning of the year, we launched Agent Hub. It is not like GetAgent where it's a question-and-answer format in the app, returning lengthy content, but leans more towards advanced users, allowing them to use command-line interfaces and process commands to execute transactions.

This direction gained some attention at that time, but the usage threshold remained relatively high. Because very few people genuinely know how to program or use command lines for trading; the vast majority of users are still regular traders who need simpler and more direct product forms.

So later, we moved the entry point to Telegram. Users just need to open a link, log into their Bitget account, and they can complete trades in a manner similar to Agent, providing a much smoother overall experience.

Cat Brother: How do you address security concerns?

Dr. Bill: For security, we have implemented sandbox isolation, four-layer identity verification, and independent environments, with the core goal of ensuring user asset safety. Additionally, we try to lower the usage threshold for regular users. Many similar products require users to connect models, manage token costs, and select service plans, which is too complex for most people. We aim to encapsulate that underlying complexity and make it easier for users to get started.

Underlying Logic of Bitget AI Trading Products and User Experience Design

Cat Brother: Which large model are you using?

Dr. Bill: We utilize multiple large models and perform intelligent allocation based on different tasks, balancing cost and effect. Simple tasks cannot use the most expensive model continuously, while complex tasks should not rely solely on cheaper models, so we are essentially conducting overall optimization.

From the product design perspective, our initial goal was to lower the entry barrier. For example, by providing users with a certain amount of free quota, and charging once that quota is used up, it's easier for users to get started. Users also don’t need to buy tokens, choose models, or can directly use the underlying capabilities we've refined.

Later, we migrated many capabilities to Telegram, including information acquisition, analysis processing, and some basic trading strategies. The product on Telegram is called GetClaw. This allows users to interact with the system as if they were chatting, providing a smoother experience. Because previously, many users actually couldn't even find the entry point when it was in the app, but in Telegram, the pathways are much more direct.

Once this experience was integrated, GetClaw quickly took off. We also conducted trading competitions to provide users with trial funds and rewards, essentially hoping to help everyone adapt to this Agent trading model more naturally.

However, we also consistently emphasize that no matter how good the tools are, trading cannot completely detach from human judgment. When to enter, when to exit, is still critical. Relying entirely on models is unfeasible, and not using them at all is also not practical. So what we want to create is not to replace users but to build tools that are sufficiently good while helping users enhance their understanding. This is why from the beginning of our AI endeavor, we've set a goal called "Empowering 100 Million Users to Stand alongside Wall Street," essentially aiming to help them become better traders.

Our goal is to make trading simpler and more personalized. For example, the system can gradually understand your trading habits, risk preferences, and operational styles, condensing the earlier complex analysis processes to ultimately present you with clear decision options. This way, when you operate, you will have more basis for your decisions and feel more secure.

Therefore, the two most core aspects of this product model are: first, long-term memory and personalized adaptation, the system can continuously learn from users; second, it is safe, effective, and the underlying tools are continually evolving. Over the past year, GetAgent has already honed many underlying capabilities, and GetClaw was built on that foundation. Of course, it’s not perfect now, and it will continue to iterate moving forward.

Cat Brother: Have you ever calculated the current trading volume for AI trading?

Dr. Bill: Right now, it’s actually still not much. Looking at the company's overall trading volume, the proportion driven entirely by AI is still very low. Because for users to trust the idea of "AI-guided trading" on a large scale, it requires a nurturing process.

Additionally, this field changes very rapidly. Large models are continuously iterating, and often the earlier product forms do not require major changes; simply switching the backend model from an old version to a new version can markedly enhance overall performance. This shows that the model capabilities and application layers have started to decouple; when the underlying model upgrades, the upper-tier experience follows suit.

So the current state is that the previous applications are rapidly iterating while the underlying models are also making continual progress; the entire ecosystem is evolving very quickly. A requirement that used to take one or two months to implement can now go live in a few days or even a day.

In this situation, what truly matters is not just development capabilities but also understanding the business itself, especially regarding trading. Because tools and models are evolving, the ultimate determinant of product value is still your understanding of the scenarios.

Competitive Advantages of Bitget AI Products and Continuous Optimization Directions

Cat Brother: Now, it’s not just Bitget; Binance, OKX are also working on AI-related products. Have you seen the skills or products they have released? What advantages do you think Bitget's AI products have compared to other exchanges? In what areas will you perform better?

Dr. Bill: That’s a great question, and we have always been very attentive to the latest developments in the industry. In terms of AI, all exchanges are at the same starting line, so we view it as an opportunity for "curve overtaking." Furthermore, AI is a field that has seen significant investments of both talent and capital, destined to become a battleground for a few leading exchanges, and Bitget's investment in this area is substantial.

In fact, since we started working on GetAgent last year, we've been exploring how AI Agents should be implemented in the crypto space. At that time, there were virtually no ready references; we could only look at how other fields approached it while continuously experimenting with our own business. Now after over a year, we have accumulated quite solid underlying capabilities along with a method for continuous iteration.

If comparing with other exchanges, I believe our advantages mainly comprise several aspects.

First, iterative experience. From the start of AI Agent development in March last year to now, we have gone through multiple quarters of continuous iteration. This process is quite painful, often feeling like rebuilding from scratch, but because of that, the experience gained can be quite profound. We can't say for sure we are the industry leader in this aspect, but at least we started early and have gone quite deep.

Second, security. When these types of Agent products first emerged, many rushed to try them but later retreated due to security issues. We always place a high priority on security internally, even if it affects development efficiency, safety must come first. After several quarters of refinement, we have yet to encounter any significant issues in AI trading or AI Agent, which is also a significant advantage.

Third, we move quickly to keep up with new product forms. Whether it’s Agent Hub or the subsequent GetClaw, we launched relatively early, and not only focused on the products themselves but also designed gameplay combining trading scenarios. For instance, we previously tried to combine AI traders with copy trading systems, allowing users to select following trades based on AI trader performance, which was an innovative step in the trading scenario.

On the surface, it seems anyone can create such products now; using some development tools, they can quickly put one together. However, the real difference comes down to how fluid, stable, and reliable the final product is; this goes beyond just which model was used, it hinges on whether you can integrate models, costs, quality, safety, and user experience effectively.

Particularly in consumer-facing scenarios, cost control is very crucial. Without optimization, the costs of such products can easily spiral out of control. Therefore, what we are doing now is not just about "which large model to use," but how to make deeper combinations and adjustments to multiple capabilities while ensuring experience and quality, keeping costs within a reasonable range.

In conclusion, I believe our advantages mainly boil down to three points: first, we started early, iterated for a long time, and have accumulated deep experience; second, our security system is relatively robust; third, in the integration of skills and product capabilities, we have developed certain methods and foundations.

Of course, if there are areas that require continual improvement, I believe the most important thing is not to keep an eye on competitors but to learn more from users. Because AI trading ultimately isn't about who offers the most functions, it's about who better understands the user. We still need to keep researching what users' perceptions, habits, and expectations regarding AI trading are today.

At the end of the day, users come to trading platforms aiming to make money. We can't guarantee users will always profit, but we hope to help them trade faster, more conveniently, and more comfortably. For example, the system ultimately only presents a few clear, personalized options and clearly explains the logic behind them, making it easier for users to make judgments and decisions with greater confidence than before.

Thus, this journey is far from over. Our current focus is on continuing to enhance the experience to be smoother, safer, and more personalized while also continuing to learn from peers and users.

Will AI Trading Lead to a "Winner Takes All" Scenario? Will Strategies Quickly Become Ineffective?

Cat Brother: You just discussed a rather ideal "AI + trading" scenario. I have two more detail questions.

The first question is that the model capabilities executing AI trading will surely vary in strength. Will there be a situation in the future where "winner takes all"? For example, those with more capital can invest in stronger large models, having more computing power and faster speeds, eventually allowing a few individuals to outperform the majority and seize all the money in the market.

The second question is that trading markets change rapidly; a set of strategies is often only effective during a certain period and can quickly be mimicked, followed, or even targeted. Does AI trading also face this limitation? Is it impossible to maintain a consistent advantage over the long term and must continuously iterate?

Dr. Bill: These two questions are indeed topics of considerable concern within the industry.

First, about "winner takes all." I think it’s unlikely to occur. One can draw parallels with the stock market, where quantitative and hedge fund industries have been developed for many years, yet even today no company has been able to seize all market profits. Even though top institutions are very strong, many participants will still exist in the market for the long term.

The reason is simple: trading systems are inherently too complex; the outcome isn't determined by just a few variables; there could be thousands of variables behind it, along with various unexpected events and black swan occurrences. Therefore, I do not believe anyone can genuinely dominate the market completely.

Regarding the second question, I feel AI trading will undoubtedly face a ceiling. Assuming perfect trading scores 100, today AI might achieve around 90, and perhaps in the future can approach 99, but it’s very hard to truly reach 100.

Cat Brother: Are you referring to a current score of 90, or will it always be limited to that level in the future?

Dr. Bill: I’m saying it’s approximately 90 today. It will continue to improve in the future, but I find it always difficult to hit a perfect score. Because the most challenging aspect of financial trading ultimately still revolves around human nature. As long as there are people behind the market making decisions, emotions, biases, and irrationality will inevitably exist.

Of course, a more extreme situation could emerge in the future where Agents, rather than humans, primarily conduct trading in the market. That would change the dynamics, as machines will exhibit much more stable discipline than people, and it would come down to model capabilities, system capabilities, and speed.

But from the perspective of the current crypto market, we are far from reaching that stage yet. Therefore, overall, this remains a continuously evolving process. As long as there are human participants in trading, it's impossible to eliminate uncertainty completely.

Cat Brother: I completely agree with your response. Because trading often relies on rationality to overcome emotions. If in the future, it’s all AI conducting trades, the competition may indeed come down to the level of intelligence and speed.

Dr. Bill: Yes, we are still quite far from that point, so there’s a lot of space and interest left in this field.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。