Written by Dou Wenxue

Edited by Ziye

On September 22, at the QianWen Office session of the Cloud Town Conference, Chen Yusen, the newly appointed CEO of DingTalk, took the stage with a microphone adorned with the QianWen Office mascot. This marked his first public appearance 100 days after assuming office. Notably, he had also taken on an additional title—Vice President of Alibaba Group.

This young executive achieved a rapid promotion from P10 to P11 within just four months. Chen's promotion is merely a microcosm of Alibaba AI's swift organizational adjustments.

Besides Chen Yusen, there were many new faces at the Cloud Town Conference, including Liu Dayiheng, who was just announced as the project leader of Foundry Qwen LLM on the eve of the conference, Zheng Bo, the technical vice president of the ATH business group, Gao Hui, the vice president of Pingtouge Semiconductor, and Li Feifei, the CTO of Alibaba Cloud Intelligence Group.

Most of these new faces are emerging leaders who have been promoted amid organizational changes within Alibaba's AI technology department and now shoulder their respective missions in Alibaba's full-stack AI strategy.

2026 Cloud Town Conference. Image source: Official WeChat Public Account of Alibaba Cloud

Today, Alibaba AI has formed a complete execution chain from foundational models to chips, cloud computing, and then to AI applications.

At the foundational model layer, Liu Dayiheng oversees the overall development of the Qwen large model; Zheng Bo is responsible for the native full-modal unified model and intelligent agent experience, leading the team along the path of multimodal unification and world model understanding.

At the chip layer and cloud and infrastructure layer, Gao Hui, the vice president of Pingtouge Semiconductor, is responsible for the research, development, and iteration of the Zhenwu series AI chips. Li Feifei, the CTO of Alibaba Cloud Intelligence Group and a member of the group's technical committee, is responsible for Alibaba Cloud's technical strategy and AI cloud infrastructure construction, leading the Agentic Cloud direction.

The AI application layer is divided into two main lines: ToB and ToC. Chen Yusen manages both DingTalk and QianWen Office, responsible for the research, development, and operation of AI-native work platforms for the B-end. Wu Jia serves as the president of the QianWen C-end business group and CEO of Quark, overseeing C-end products such as the QianWen App, Quark, AI hardware, UC, and Shuqi.

However, at this critical juncture where AI is shifting from technological narratives to commercial realization, the burden on Alibaba's technical executives is heavy.

Chen Yusen needs to prove that Alibaba has not fallen behind in the AI office wave; Wu Jia must demonstrate that the QianWen App can serve as a super entrance and attract C-end users; Liu Dayiheng must drive the foundational model to continuously catch up with the industry's top tier.

Zheng Bo needs to translate the technological advantages of multimodal and world models into usable products; Gao Hui must continue to advance the mass production and ecosystem construction of the Zhenwu chips; Li Feifei must complete the engineering of Agentic Cloud to enable cloud infrastructure to meet the practical requirements for large-scale deployment of intelligent agents.

In addition to external competition, internal competition is also underway, especially between QianWen Office and QianWen App, both vying in the AI office sector. Now, both are emphasizing the importance of "context," and the competition for talent between the two products is no secret in the industry.

In the AI era, with executive adjustments, internal resources and power are being redistributed. Whether the various gears of Alibaba's AI landscape can mesh seamlessly and who will truly stand out are the tests that this group of technical executives must face.

1. Chen Yusen's Triple Jump: Greater Expectations, Heavier Burden

Among Alibaba's recent executive appointments, Chen Yusen's promotion has been particularly rapid.

An entrepreneur by background, Chen founded a cybersecurity company, Changting Technology, at the age of 22. Due to its white-hat hacker team and practical attack and defense technologies, it was acquired by Alibaba Cloud.

Subsequently, Chen ventured into entrepreneurship within Alibaba Cloud, leading the development of an AI Agent (intelligent agent) marketplace and digital workforce platform—MuleRun. Within one month of its launch, the product registered over 500,000 users.

Chen Yusen truly gained widespread attention when he succeeded Wu Zhao as CEO of DingTalk in June this year. At that time, Wu Zhao, the former CEO of DingTalk, faced controversy over ineffective formalized management due to an essay written by a departed employee titled "Inside DingTalk." After Wu Zhao's departure, Chen Yusen took over DingTalk, and this post-90s executive became a focal point of industry discussion.

Chen Yusen, Vice President of Alibaba Group. Image source: Official WeChat Public Account of Alibaba Cloud

Now, with his promotion to Vice President of Alibaba Group, it signifies that Wu Yongming holds him in high regard, and QianWen Office is becoming a crucial part of Alibaba's current AI strategy.

To date, the market is watching the performance of this new executive, who carries a heavy burden.

During his early days in office, the AI office market was vying for desktop entrances for AI Agents. According to data released by Analysys, in June this year, Tencent's WorkBuddy recorded 20.97 million monthly visits, while ByteDance's TRAE IDE (domestic version) recorded 12.79 million. In contrast, Alibaba's QoderWork and Wukong combined recorded only about 9.19 million, roughly one-third of Tencent's figure, ranking last among the three giants.

Against the backdrop of AI office integration, Chen Yusen acted swiftly.

He first integrated the Wukong and MuleRun teams into a new Wukong team and then combined QoderWork, Wukong, and MuleRun into QianWen Office.

QianWen Office also disclosed new developments almost monthly, first launching its public beta on August 3, and then releasing its first-month performance report on September 4, claiming over 30 million users, with enterprise users accounting for more than half. It completed 120 version updates in the past month, with an average of four iterations per day.

However, this did not provide much of a head start. QianWen Office does not follow the low-barrier approach like WorkBuddy, nor does it offer as much freedom in large model utilization. It also lacks the support of AI applications with a large C-end user base like Doubao. DingTalk, which has historically focused more on decision-makers, does not meet user demands for daily office work as closely as Feishu.

Furthermore, the disclosed data refers to registered users rather than monthly active users, retention, or paying users, casting doubt on its value.

Currently, the direction of competition in the AI office market and even the personal AI assistant sector has shifted again.

As the model capabilities of major companies gradually converge, what determines how far an Agent can go is no longer the parameter scale but the amount of effective information it can mobilize about the enterprise, business, or individual during reasoning. In other words, contextual engineering capabilities are becoming the core of the current AI office sector.

Chen Yusen directly set the theme of his speech at the Cloud Town Conference as "Context is All You Need." On the same day, QianWen Office launched a new enterprise-oriented product, "Enterprise Context," which primarily connects, structures, and compresses documents, meetings, group chats, and even tacit knowledge accumulated by enterprises over the years, providing them to Agents performing tasks on demand.

Coupled with Alibaba's full-stack AI advantages and its high ecological position within Alibaba, such as key meetings attended by group executives like Wu Yongming, resources are also tilting towards QianWen Office. For example, the flagship model Qwen3.8 was released and first integrated with QianWen Office on the same day, while the QianWen App adopted the same model four days later.

However, this does not guarantee a smooth future for QianWen Office.

Achieving true contextual capabilities is not easy. QianWen Office aims to build independent enterprise contextual infrastructure, requiring capabilities such as long-text and long-term memory accumulation, multimodal contextual correlation, and cross-application and heterogeneous data integration. It requires not only model understanding but also enterprise and business understanding, as well as enhanced security.

Image source: Official WeChat Public Account of Alibaba Cloud

Competitors are also acting swiftly. At the Feishu Future Infinity Conference held a week before the Cloud Town Conference, Feishu CEO Xie Xin also emphasized "context" in his speech, stating that Feishu provides context, while Doubao Work transforms it into productivity.

A more direct challenge comes from within. The QianWen App is also targeting AI office as a commercial breakthrough, accelerating its evolution towards Personal Agents (personal intelligent agents), with the core being to accumulate broader personal context on its products.

These two applications, both named "QianWen," may seem to have clear divisions of labor—one for ToB and the other for ToC—but for the nascent AI office market, such naming confusion may lead to mutual hindrance in product promotion.

It is no secret that the two products are competing for talent.

An interesting scene was reported by the media: On the afternoon of September 22, the main agenda of the Cloud Town Conference focused on the QianWen App, with core speakers including Zheng Sishou, the product leader of the QianWen App, and Song Gang, the general manager of QianWen AI Hardware. The sub-forum for QianWen Office was held simultaneously with the main forum but at a location 15 minutes away. One hour before the meeting, the QianWen Office team visited the media room and took away most of the journalists who were still having lunch or resting.

Behind this scene, both QianWen Office and QianWen App are at a stage where they must prove themselves, whether in external competition or internal competition.

2. QianWen App Follows the Trend, and Wu Jia is Also Anxious

Compared to QianWen Office, the QianWen App and Wu Jia may be even more anxious.

Over the past six months, the direction of the QianWen App has been constantly changing.

In January this year, as internet giants vied for the super entrance during the Spring Festival, the QianWen App became Alibaba's new attempt at an AI super entrance after Taobao, carrying high hopes.

The vision for the QianWen App was to transcend its chatbot nature and upgrade into an AI life assistant capable of "getting things done" for users. In a short period, it integrated with Alibaba's ecosystem businesses such as Taobao, Alipay, Taobao Deals, Fliggy, and Gaode, launching over 400 functions. Beyond Alibaba's ecosystem, the QianWen App also opened up to third-party Agents, with merchants including SF Express, Tianya Daojia, and Shansong.

Additionally, the QianWen App attracted users with substantial subsidies during the college entrance examination period, reaching a peak of 70 million daily active users.

However, becoming a super entrance is no easy task. The QianWen App's entrance and ecosystem story did not translate into user retention. The decline in DAU forced the QianWen App to seek new directions.

We have seen that the QianWen App, without prior foundation, began exploring office scenarios and introducing fees. It launched several features such as thinking and research, scheduled tasks, office assistants, voice calls, and an Agent plaza, while integrating the latest flagship model Qwen3.8-MAX. Among them, the "office assistant" became a paid item, with three pricing tiers: Premium, Elite, and Flagship, priced at 19 yuan/month, 49 yuan/month, and 128 yuan/month, respectively, with a maximum annual subscription of 1,499 yuan.

Now, the QianWen App is telling the story of Personal Agents, officially stating that it can connect users' memos, calendars, emails, and other applications, operating computers and browsers to complete multi-step tasks such as searching for information and downloading attachments.

Image source: Official WeChat Public Account of Alibaba Cloud

It is evident that over the past six months, the QianWen App has been closely following industry trends, trying whatever is popular in the market. However, its current Personal Agent story is at a stage where the strategy has been announced, but user perception is weak. What advantages it can form in the industry remains to be seen. Like its office story, the Personal Agent story is conceptually appealing but lacks outstanding product strength.

Given the rapid changes in AI technology trends, it is understandable to explore new directions by following trends. However, it does give the impression of a fluctuating positioning, leaving outsiders unsure of the QianWen App's core competitiveness.

Moreover, it is not easy for an AI application to excel in multiple areas with short-term efforts.

The QianWen App has long faced the issue of user loss without promotion, rooted in its lack of truly compelling features to retain users.

Alibaba is unlikely to invest billions in subsidies for the QianWen App again, as it did during the Spring Festival promotion. In the first quarter of fiscal year 2027 (ending June 30, 2026), the high investment in the QianWen App was cited as the main reason for the significant losses recorded in Alibaba's "AI Labs and Applications" division, which reported revenue of 3.338 billion yuan but an adjusted EBITA loss of 13.861 billion yuan, approximately 4.2 times the revenue.

For Wu Jia, two core challenges lie ahead: finding scenarios for the QianWen App that can truly retain users and enabling the QianWen App to achieve self-sufficiency and complete the crucial leap towards commercialization.

From current explorations, it is not easy for the QianWen App to overcome these two challenges.

In the office scenario, the QianWen App has little chance of success in its competition with QianWen Office.

Theoretically, the QianWen App focuses on personal context and lightweight office work, following a route more similar to Doubao + Feishu's contextual approach, targeting C-end personal efficiency scenarios, and differing from QianWen Office's enterprise-level B-end positioning.

However, the ultimate operators of office systems are still humans, leading to significant overlap in functionality and user demographics between the two at the implementation level.

More critically, the QianWen App lacks the long-term accumulation of personal user habits and AI scenarios, making it difficult to form stable user loyalty and engagement. Its connection with the DingTalk ecosystem is also weaker than that of QianWen Office, which is natively bound to DingTalk. DingTalk's past focus on decision-makers may also align better with QianWen Office's enterprise context route.

Image source: Official QianWen Website

Therefore, even though the QianWen App's pricing appears lower than both Doubao and QianWen Office, its new story has not gained much traction based on the current situation.

3. New Generation of Technical Leaders at the Helm, Alibaba is Still Laying the Foundation for AI Applications

At the main forum of the Cloud Town Conference, Wu Yongming proposed in his keynote speech that "AI models, AI chips, and AI cloud are the three cornerstones of the machine intelligence era," indicating the direction of Alibaba's long-term strategic investment.

In fact, Alibaba has been telling the story of full-stack AI for many years, and Wu Yongming's attitude remains unchanged today, except that his executive team is now more complete.

Over the past six months, Alibaba's technology department has been undergoing organizational adjustments, with some of today's technical leaders gradually emerging amid these changes.

Image source: Official Cloud Town Conference Website

In March, Alibaba formally established the Alibaba Token Hub (ATH) business group, directly led by Wu Yongming, encompassing the Tongyi Laboratory, MaaS business line, QianWen business unit, Wukong business unit, and AI Innovation business unit, covering the complete chain from foundational model research and development to AI applications for individuals and enterprises.

At this time, the Qwen team, which had lost its leader Lin Junyang, was also undergoing internal restructuring. Liu Dayiheng, previously responsible for pre-training, began to oversee post-training and the Coding team, reporting to Zhou Jingren.

In April, Wu Yongming issued another internal letter, announcing the establishment of a technical committee at the group level and upgrading the Tongyi Laboratory to the Tongyi Large Model business unit.

In this internal letter, Li Feifei, formerly the president of the Database Product business unit, officially took over the reins of Alibaba Cloud's technology from Zhou Jingren, the former CTO of Alibaba Cloud.

In June this year, Alibaba announced the merger of the Tongyi Large Model business unit and the Future Life Laboratory to form the Token Foundry business unit, directly led by Wu Yongming. Alibaba Group Vice President Zheng Bo led teams such as the video generation large model Happy Horse and the open-world model Happy Oyster to join this business unit, becoming a pioneer in this integration.

Besides these newly appointed executives, there is also Gao Hui, the Vice President of T-Head Semiconductor, Alibaba's chip subsidiary. These executives will all become the main force behind Alibaba's full-stack AI strategy.

Today, these executives have all achieved some success in their respective fields.

Liu Dayiheng drove Qwen3.8-Max to autonomously complete 33 rounds of iteration with nearly “zero human participation”, improving evaluation scores by 12.5%, reasoning throughput by 96%, and reducing the physical footprint in hardware collaboration by 42%.

HappyHorse-1.0, developed under Zheng Bo’s leadership, made a surprise appearance on the Artificial Analysis blind-test platform in April 2026, topping both the text-to-video and image-to-video categories simultaneously.

The Zhenwu chip, overseen by Gao Hui, has delivered commercial results. As of April 2026, cumulative shipments of the Zhenwu chip reached 560,000 units, serving over 20 industries including finance, intelligent driving, internet, energy, and manufacturing, with more than 400 clients.

After Li Feifei took over as CTO of Alibaba Cloud, Alibaba Cloud launched a series of new products such as AgentCore, Agent Sandbox, and a new generation of high-performance storage CPFS. Among them, CPFS provides throughput at the hundred-TB/s level, reducing the average model startup time by 50%, increasing peak computing utilization by 30%, and lowering AI storage costs by 69% in actual model training.

Judging from the situation at the Yunqi Conference, several executives are already planning for the future.

For example, Liu Dayiheng released five roadmaps for Qwen at the Yunqi Conference: self-improvement, architectural iteration, unified multi-modality, open-source and openness, and on-device and endpoint deployment. Zheng Bo plans to integrate image, video, and music generation with real-world understanding within three years, completely abandoning patchwork solutions, while also announcing that a new video model will be released in November, and so on.

However, the executives are also under significant pressure.

For Liu Dayiheng, foundational models have entered a stage of incremental competition. The capability gaps between various industry models continue to narrow, making breakthroughs based solely on evaluation metrics increasingly difficult. On one hand, he must continue to invest in computing power for model iteration and maintain the vitality of the Qwen open-source ecosystem. On the other hand, he must balance R&D investment with return on investment, delivering on the roadmap goals of self-iteration and unified multi-modality.

Zheng Bo faces fierce competition in the generated video space. Current text-to-video products generally suffer from shortcomings in visual coherence and physical realism, and the industry still lacks mature solutions capable of large-scale commercialization. Training video large models consumes enormous computing power. He must deliver on the long-term goal of integrating audio, video, and real-world understanding while finding differentiated scenarios in competition with leading rivals to avoid models remaining stuck at the Demo stage.

Gao Hui faces the ongoing challenge of commercializing AI chips. The domestic AI chip sector is crowded with competitors, and customers prioritize stability, cost-effectiveness, and complete software-hardware adaptation when making selections. Although the Zhenwu chip has achieved large-scale shipments, it must continue to expand into automotive and government/enterprise clients, iterate chip architectures, keep pace with the growing computing demands of large models, and withstand pressure from both overseas established chip manufacturers and domestic chip competitors.

Image source: Yunqi Conference official website

Li Feifei’s core challenge is balancing the commercialization of cloud and AI infrastructure. High costs for AI training and inference storage are industry-wide issues. Infrastructure products like AgentCore and CPFS need to convince enterprise clients to pay. She must continue refining cloud-native AI products to lower AI usage costs for clients while safeguarding Alibaba Cloud’s core cloud business.

From a group-wide perspective, these executives share an additional layer of pressure.

Alibaba’s full-stack AI chain is long, with interlocking links from chips, cloud, and foundational models to C-end and B-end applications. Technical goals vary across lines, easily leading to fragmented team efforts and internal resource competition.

Executives must not only win external competition in their respective domains but also collaborate across businesses to smoothly transmit foundational technical capabilities to upper-layer AI applications, ultimately proving that this full-stack layout can achieve a viable commercial closed loop (closed loop).

(The header image of this article is sourced from the official website of Alibaba Group.)