Alibaba's campus was enveloped by AI-ordered milk tea. On January 15th, Wu Jia, President of the QWEN C-end Business Group, stood on stage and spoke into the QWEN App: "Order 40 cups of Bawang Tea Ji's 'Bo Ya Jue Xian' for me." QWEN directly placed the order and completed the payment. Shortly after, Taobao Flash Delivery riders arrived at the scene with the milk tea. There was no need to switch between apps or toggle backgrounds; everything was accomplished seamlessly within the QWEN App itself. This demonstration reflects a top-down strategic will. Alibaba is attempting to use AI as a lever to reshape the formerly "fragmented Alibaba" into a highly efficient and synergistic "new Alibaba kingdom." At the event, Wu Jia announced that the QWEN App is now fully integrated with Alibaba's core ecosystem businesses, including Taobao, Alipay, Fliggy, Amap, and Taobao Flash Delivery. The AI, which was once only capable of chatting and creating images, has suddenly gained a range of capabilities like ordering food delivery and booking flight tickets. "QWEN is the first AI that can truly help you get things done," Wu Jia stated. He expressed that the era of AI handling tasks has just begun, and his goal is to build the QWEN App into the most powerful human AI assistant, aiming to cover every individual in the future. This is not merely a functional iteration of an app; it resembles a "self-reconstruction" of Alibaba for the AI era. Alibaba's ecosystem is vast, spanning from shopping and food delivery to maps and travel. However, creating a synergistic effect for the end-user has been a challenge; during the mobile internet era, it instead seemed to scatter the entry points. If QWEN can truly become the "most powerful human AI assistant," Alibaba will achieve a leap from "shelf e-commerce" to "command e-commerce." This would not only address Alibaba's weakness in having a high-frequency C-end traffic入口 but also build a closed loop centered around itself. Users could ask for directions, hail rides, order food, and shop all within QWEN, keeping all data and transactions within the Alibaba system and no longer relying on external traffic feeds. However, this is no easy task. The logic of "shelf e-commerce" is to let users browse, offering massive choices with high tolerance for imperfection. In contrast, the logic of "command e-commerce" is to choose for the user, pursuing precision with an extremely low error tolerance. Having AI "handle tasks" essentially means making the AI承担 decision-making responsibility. This places higher demands on the reasoning capabilities of Alibaba's large model and the real-time responsiveness of its service supply chain. The QWEN C-end Business Group was born into fierce competition. In a post-event interview, Wu Jia candidly admitted that there might only be two or three final players remaining in the endgame. Nevertheless, in his view, the old internet logic of burning money and buying traffic to win the market has become失效. Only companies that can genuinely elevate their models' intelligence level and have the resources to invest heavily in their ecosystem will earn a seat at the final table. Major players with this qualification are exceedingly rare. Thus, under Wu Jia's leadership, a "second startup" for Alibaba has vigorously commenced. Whether QWEN can become the "J.A.R.V.I.S." in every Chinese person's phone remains an unanswered question. What is certain is that when AI starts making payments for you, a new commercial era has begun. In this era, whoever can first make AI "handle tasks flawlessly" will be the new king.
The following is an excerpt from the dialogue with Wu Jia, President of the QWEN C-end Business Group: Question: What are the core directions for the iteration of泛智能 and QWEN in the next six months? Wu Jia: In the next six months, integrating into the Alibaba ecosystem, expanding the boundaries of task-handling capabilities, and strengthening the model's understanding能力 in life scenarios are our very important main focuses. Achieving universal satisfaction in life scenarios is quite difficult; consistency is better in office and learning scenarios. We still want to leverage the rich supply of the Alibaba ecosystem, combined with our model's capabilities and our understanding of user needs, to build a life-scenario product that is globally leading. Question: How does QWEN balance the conflict between efficiency and depth of thought? Wu Jia: We have an internal term: 'appropriate.' I don't think AI should equal extreme simplicity. For example, if I want to write a research report, I don't want the AI to give me a first draft but to co-create with me. Communication between AI and humans is essential; it's the same in life scenarios. Sometimes you don't even know your own real needs, so you need the AI to communicate with you, even proactively offer suggestions. I believe the key to AI is intelligence, not efficiency. It's just that at this current stage, intelligence manifests more in improving efficiency. But higher-level intelligence isn't solely about efficiency; AI needs to think like a human. Currently, we are更多的是 using a model to提升 its intelligence, improving satisfaction across different needs. Question: QWEN is already the unified intelligent entry point for the Alibaba ecosystem. What about the current collaboration architecture, resource allocation, and coordination challenges with various departments within the Alibaba Group? Wu Jia: In the AI era, we won't undertake many separate tasks specifically to integrate a particular service. It's essentially one model接入这么多 services. Each Business Unit just needs to register its tool capabilities with the AI. During this process, we need debugging. The current method involves us forming a common virtual team with all second-party units within the group. This is a win-win situation. The bigger QWEN grows, the more it creates增量 in life services. The大量 new services generated by AI in the future will also be增量, not存量. Question: Will QWEN only integrate with the Alibaba ecosystem? Wu Jia: No, we will open up official cooperation after the Spring Festival. But we will choose a specific timing. As everyone can see, whether internationally or domestically, many companies use different methods. How do we achieve a good technical approach? It also needs to integrate with the domestic Chinese ecosystem, as the international ecosystem differs from China's internet ecosystem. So, we need to determine what method we will use. Additionally, regarding the model's usability and convenience, integrating the Alibaba ecosystem can also help us summarize better methods. But directionally, we will definitely open up. Question: In terms of capability, roughly how long will it take for QWEN to cross over from being a novelty to a daily necessity? Wu Jia: We are currently seeing quite good retention rates. I believe the functions we released today are all user necessities; we haven't focused heavily on偏娱乐 or creative directions. We do work on them but haven't invested as much effort. We invest more effort in necessities because办公 is ongoing, learning is ongoing, ordering food delivery is ongoing – these are the areas we focus on. So, the retention I see is acceptable. If someone churns, we should do better. Question: What is the logic behind your decisions on what to do and what not to do? Wu Jia: From a demand perspective, we currently focus on high-frequency, essential needs – that's certain. Secondly, we focus on the capabilities that AI can deliver today, especially for the C-end. I think the scope of AI's capabilities is also acceptable; it can't do everything. But we are currently in a leading position in the Chinese market, so relatively speaking, in terms of choices, we haven't explicitly ruled anything out. I think many questions today still focus on the difference between developing AI products and traditional products. Previously, for a traditional product, we would decompose it into dozens of projects, allocating resources here and there. But now it's different. The model is there, and 90% of the product might already achieve 80% satisfaction – that's where we spent 70% of our effort. We aren't specifically optimizing one thing; we are提升 coding ability, execution ability, planning ability, etc., all at once. So, we have a very important approach: we abstract user needs based on high frequency, essentiality, model capability, and the Alibaba ecosystem. We translate these abstractions into directions for model iteration and Agent iteration, and then we proceed. For particularly long-tail needs, some things are indeed very difficult, even for humans to handle, so we wait for the next stage. Question: How do we further leverage our ecosystem advantage and cultivate it meticulously? Wu Jia: We currently have three lines of focus. The first main line, the long-term line, is still the model and the Agent – this is about shoring up weaknesses and is long-term. Looking at the market today, even ignoring AI products, and just considering all current products, we have a ranking of satisfaction for this type of user需求. Combining these two lines, we will iterate in layers. Basically, there will be a major model iteration every quarter. This version we will develop together with Tongyi – this is a very important line. Based on this, the提升 of Agent capabilities requires some follow-up work – this is another line. Then we look at the third line. So, these three lines will be promoted with relative节奏. Generally speaking, I believe the core drivers today are still technology, data, and the ecosystem. It's not an iteration model focused solely on fixing bad cases, because we are still in a rapid phase of capability acceleration. So, from this perspective, we hope that in our next version, for instance, our next version of the life assistant, will focus more on personalized operation. But translating that statement involves model-related issues. However, we might not say the next version will make ordering food delivery exceptionally good – we will definitely work on it, but some of its capabilities we will abstract and work on separately. Food delivery has some development-focused functions, like what was just mentioned: can the pickup code be displayed on the order, etc. These are things at the experience level. Question: So how does Alibaba assess the impact of QWEN on its existing retail or e-commerce businesses? Wu Jia: We haven't yet observed a situation where people open QWEN and then stop opening Taobao. I believe we will create增量 because it's more convenient, the barrier is lower, and new habits will create增量. However, I think we also cannot排除 the possibility that some people will eventually get accustomed to ordering food delivery through QWEN instead of traditional platforms; they will develop a new habit. People aren't that纠结 about it; they look at whether the user's frequency of visits has increased and whether the session duration has improved. Question: Are the current iteration goals for the model at QWEN completely different from those of the base model team? Wu Jia: Our iteration goals are a subset of the base model's iteration goals; they encompass ours because we need to build our own business on top of the base model. Of course, for some application-side aspects, we will also conduct some post-training on the model, but we do so on top of its model. Our major version releases, which I mentioned occur every 3 months, will basically feed the demands observed over those 3 months to the base model team. The base model team will also help us update a version. So, we are a subset of their iteration targets because the base model serves the entire Alibaba ecosystem. Question: If developing a general AI assistant application involves engineering on one hand and the base model on the other, which area currently offers higher efficiency gains through work? Wu Jia: Many people online are discussing this topic: whether more data makes a model better, or if the C-end doesn't need that much intelligence. I believe that today, at least within my scope, Chinese businesses have moved past that stage – we all operate with one model, not many models. It's not about having a small model for areas requiring low intelligence and a large model for high intelligence. They are all large models – not small-sized large models versus large-sized large models, but rather a model smart enough to know that simple problems require simple reasoning to explain, and clever problems require clever model answers. So, after reducing the number of models, the interface for iteration with the base model becomes clearer. Secondly, is data capability important? Of course, it is. Additional data is very important, especially in life scenarios. Because model training uses data from a specific time period, and China's supply is so rich, and data changes so rapidly, solving the timeliness issue of data is naturally crucial; it's core to building capability. Therefore, long-term benefits will definitely come from the development of the base model. If we look at efficiency over one or two years, the base model is undoubtedly the most important. But if you're talking about short-term efficiency, iteration speed, and feature implementation, then post-training might be more显著. Question: Within Alibaba's current AI-to-C entry points, how will QWEN and Quark be differentiated? Wu Jia: I think they are still different. Quark is an AI browser, AI search. Some people have a strong need for an AI browser. QWEN is an AI assistant; it's more like a person, so some people have a strong need for an AI assistant. The commonality is that all AI functions are functions within QWEN; the AI functions in Quark are also functions from QWEN. I think it's a matter of path. No matter how time develops, AI browsers and AI搜索 won't disappear in the future, but their proportion in the AI era might not be as large as conversation. However, they are also user habits – invoking QWEN from within the AI browser, invoking QWEN from within search. I see them as two different user-facing interfaces, but all the AI is QWEN's. So today, we see on the PC端 that users are split half and half – half prefer QWEN directly, and the other half prefer opening QWEN within the AI browser (which is still QWEN). I think we don't纠结 over this. But on the mobile端, increasingly more people will use QWEN directly. Question: There's a lot of competition for the AI入口 now. What is our strategic layout for this C-end入口? Wu Jia: The current situation is indeed like that; you see this issue on many different 'ends'. But I believe the final outcome won't have so many players, because there are very few companies that can truly elevate their models' intelligence level and have the resources to invest in their ecosystem like this. I think ultimately, only a few will be able to provide this capability, perhaps one or two, or two or three. But will the front-end interfaces disappear and consolidate into one or two? That's hard to say. Today is the early stage of AI development, so everyone feels there's an opportunity. At this stage, it's still relatively easy to create an AI assistant with a different style. People think, 'I am this style of AI assistant, you are that style, we seem different.' But actually, I believe starting from the first half of this year – I mean the first half of 2026 – it won't be like this. Because we are conducting大量 tests online now, and actually, different Agents have different styles. We might use one main style currently, but in the future, we will change. It's just like in real life: I can become friends with you, but I might not necessarily become friends with him. You might like my way of expression but not his, even if we are talking about similar things. So, I think when we reach the stage of true personalization and拟人化, many companies will stop developing AI. I believe only a few will remain in the market. Question: Is this integration of the group's APIs also because you foresee very significant improvements in both model capabilities and Agent capabilities in the next 3 to 6 months? Wu Jia: We have always believed that this direction is the core of the core for developing AI products. This isn't something we suddenly realized we needed to do; that's the first point. Secondly, we observed the trend of Agents in life aspects around July of last year, following the emergence of reasoning models, including progress in VRL. We already saw this trend back then. From that time on, we have been continuously increasing our investment. Question: The competition among major companies' Chat tools is quite fierce now, with one already having a DAU exceeding 100 million. From the perspective of DAU or MAU data, is QWEN very concerned about it? Wu Jia: Even in traditional internet terms, the difference between 80 million and 100 million DAU isn't that significant. I firmly believe in one viewpoint: in the intelligence era, the key is whether the product has crossed the intelligence threshold, whether it can truly serve and execute like a human, and whether it can achieve high accuracy and satisfaction rates in the digital world. Crossing the intelligence threshold is related to how much traffic you pour into it, but the correlation isn't as strong as in traditional internet. The decisive factors are still the AI model's training paradigm and a whole series of other factors. Therefore, for Alibaba, a company非常专心 and investing huge effort in model development, we focus on whether we have crossed the intelligence threshold. We look at the global progress in AGI. Question: So user experience is the primary consideration? Wu Jia: The world has never used something like this before, just like when people used the iPhone for the first time. When the iPhone 1 came out, the user experience wasn't that great. By the iPhone 4, it had completely changed. Today, AI might still be in an iPhone 1 era. AI making good decisions – we are working on that too, and there will be a series of methods. It won't be like the current AI pinpointing a single product precisely. I think AI replacing the traditional 'scrolling' action has a very good opportunity; just give us a bit more time. Question: Currently, AI is颠覆 traditional traffic models, including mobile operating systems and the APP commercial ecosystem. As AI becomes more prevalent and used in the future, how will the APP ecosystem evolve? Also, will the boundary between QWEN and Taobao become increasingly blurred? Wu Jia: I think it will still take some time to reach that stage. First, I don't believe there will be that many large, entry-level AI Agents in the future. With technological development, the competitiveness of comprehensive Agents today is still very strong. Agents are increasingly proven to be a阶段性 product, distinct from simply accessing a comprehensive Agent. It's somewhat analogous to the relationship between mini-programs and WeChat, or merchants and Taobao. I don't think many independent Agents will evolve as primary入口; it will be more of a comprehensive entity. From a demand perspective, All-in-One is a trend. Question: Or perhaps we can predict some of the most likely changes for 2026? Wu Jia: Regarding AI, I think it will become more and more like a person. Its way of thinking and doing things will make you feel that you are increasingly interacting with something that understands human needs.
Comments