Alibaba has finally streamlined its AI office product suite.

On August 3, Alibaba's enterprise-level AI agent product, QwenWork, officially entered its public beta phase. Both individual and corporate users can now access the web version and PC client via the official website, with the DingTalk integration set to launch soon.

Prior to this, QwenWork underwent a closed internal beta test. It was reported that the product would deeply integrate three AI agent tools from Alibaba: QoderWork (specializing in desktop-based local file operations and code assistance), Wukong (seamlessly integrated with DingTalk for enterprise collaboration), and MuleRun (focused on cross-platform task automation). Chen Yusen, the newly appointed CEO of DingTalk, is spearheading this initiative.

This move signifies that Alibaba has unified its AI office offerings under a single platform, elevating it to a top-tier strategic product within the company.

As the industry's first product to simultaneously support desktop agents, cloud agents, and enterprise collaboration agents, QwenWork is clearly positioned as an enterprise-level AI productivity platform for organizations, rather than just a personal AI office tool for consumers.

Not Directly Targeting WorkBuddy

An often-overlooked detail is that Alibaba also released its flagship model, Qwen3.8, which serves as the foundational model for QwenWork. With 2.4 trillion parameters, this model further enhances capabilities in coding and collaborative work.

To some extent, office scenarios represent the ideal "training ground" for large AI models. Compared to chatbot-style Q&A interactions, office environments provide a more consistent and high-intensity volume of token usage.

As QwenWork becomes integrated into office workflows, daily use by enterprise users translates into repeated model calls, generating valuable data and feedback. The more the model is used, the smoother its performance becomes, creating a dependency on the workflow. This effectively positions the model within high-frequency workflows, securing its relevance. If QwenWork achieves significant adoption at the enterprise level, Alibaba will have another product driving token circulation, setting a new benchmark for evaluating the value of its AI investments. The market has already responded positively; as of the lunch break on August 3 in the Hong Kong stock market, Alibaba's stock had surged by over 7%, with its total market value recovering to approximately HK$2.4 trillion.

There's also a longer-term perspective to consider. Ultimately, model competition hinges on iteration speed, which depends on the quality and density of feedback. Feedback from consumers tends to be fragmented and prone to inaccuracies. In contrast, feedback from office scenarios directly correlates with task completion, making success or failure immediately apparent. The value of such feedback for model improvement is on a different scale.

This raises a more intriguing question: Who is Alibaba targeting with this product?

For many, the immediate assumption is Tencent's WorkBuddy, which currently leads in consumer-facing AI office solutions. Positioning QwenWork against WorkBuddy seems like the most straightforward narrative. After all, targeting a visible and accessible competitor makes for an easy story to tell.

However, internally at Alibaba, the business logic differs. Huxiu revealed that in May and June, the business team and Alibaba's decision-makers held multiple discussions to define QwenWork's positioning. The consensus was that Alibaba's opportunity lies in the "enterprise-level" market.

This detail is worth pondering. The fact that a product's positioning required repeated deliberations between the business team and decision-makers suggests genuine internal debate.

It's not hard to imagine that the allure of consumer-end traffic caused some hesitation within Alibaba.

Tencent's AI efforts, once criticized for being slow, seem to have found their stride with WorkBuddy, breaking through previously perceived barriers.

According to data from Analysys' "China Office Intelligent Agent Platform Market Research Report 2026," in June 2026, monthly visits to 17 mainstream desktop-based AI-native office intelligent agent platforms in China exceeded 60 million.

Among them, Tencent's WorkBuddy ranked first with 20.97 million visits, surpassing the combined total of the second and third-ranked platforms. ByteDance's TRAE IDE (domestic version) came in second, while Alibaba's QoderWork ranked third.

Additionally, Tencent's Q1 2026 earnings report disclosed that WorkBuddy's active user retention rate exceeded 60%, while its paid user retention rate surpassed 80%.

The temptation was significant, but Alibaba ultimately made a strategic choice to focus on the enterprise-level AI office market.

In reality, if QwenWork were to directly compete with WorkBuddy or TRAE IDE, it would essentially be fighting on the opponent's home turf.

This approach is all too familiar to Alibaba, akin to Taobao's flash sales taking on Meituan—a war of attrition based on traffic and advertising spend. While Taobao achieved its strategic goals, the cost was immense, with nearly RMB 100 billion in group profits eroded.

Alibaba's decision-makers believe that the AI office market is still in its infancy, with the final outcome far from decided. WorkBuddy's current lead is only temporary, and Alibaba still has room to expand its advantage.

Enterprise-Level AI Office: Alibaba's Strategic Focus

This judgment is a measured one.

While consumer-facing AI office products are more visible, with download numbers, engagement rates, and social media buzz easily quantifiable, the real revenue and barriers in the AI office business lie in the B2B sector.

Whether individual users are willing to pay remains unproven. However, AI rankings indicate that Doubao lost 18.65 million monthly active users from May to July, likely due to the introduction of its paid version, which drove many users to other AI applications.

Instead of engaging in a protracted battle with WorkBuddy in the consumer market, Alibaba has shifted the battlefield to enterprise-level AI productivity platforms. Its goal is not just to win with a single product but to secure the entry point for the next generation of office infrastructure.

What truly sustains an office product are organizational relationships, permission systems, data security, and integration with enterprise business processes—elements that consumer-end traffic cannot replace as a competitive moat.

In my view, there are three main reasons why QwenWork is positioned as an enterprise-level AI office platform:

First, the demand for enterprise-level AI office solutions is just beginning to surge. Enterprises are transitioning from observing AI to actively adopting it. Cost reduction and efficiency improvement, once just annual report objectives, are now being broken down into specific roles and processes. Moreover, AI office represents a "dimensionality reduction" attack on traditional SaaS models, as enterprises are purchasing the ability to directly deliver results.

Take Feishu, which recently underwent adjustments, as an example. Over the past six months, Feishu has struggled under the pressure of AI office adoption. Its core products—Feishu Aily, Feishu Miaoda, and the Multi-Dimensional Table Agent—along with supporting tools like the OpenClaw official plugin and Feishu CLI, remain embedded within Feishu's original framework.

Feishu has also become increasingly bloated and complex. Xie Xin has attempted to position Feishu as China's answer to Office, Slack, Notion, and Microsoft 365 all in one.

At its core, this remains a reformist approach of "SaaS + AI"—like mounting new cannons on an old ship without altering its structure, only increasing firepower.

However, when competitors dismantle the ship and build a speedboat, no amount of firepower can match their speed or pose a direct threat.

ByteDance's internal memo on Feishu's restructuring also noted that in Q2 2026, over 90% of Feishu's new clients simultaneously purchased Feishu's AI products. This indicates that enterprise clients are not just paying for Feishu but for AI capabilities. In the age of agents, the "All-in-one SaaS" approach is no longer viable.

Furthermore, ByteDance and Tencent have only recently turned their attention to enterprise-level AI office solutions. Tencent launched WorkBuddy Enterprise Edition in June to enter the enterprise market, while ByteDance just completed its largest-ever restructuring of its ToB business, disbanding Feishu after a decade of independent operation and integrating it into Doubao and Volcano Engine.

Second, the barriers to entry for enterprise-level AI office solutions are high. Model capabilities are merely an entry ticket; the stability of the underlying infrastructure, cloud support, enterprise service experience, and the ability to handle delivery and compliance are the true measures of internal strength.

Sources familiar with the matter told Huxiu that enterprise-level AI office solutions must not only ensure robust general agent capabilities but also have deep insight into and meet enterprise-level scenario demands. In these areas, Alibaba has accumulated millions of enterprise clients through DingTalk and Alibaba Cloud, with vast resources in demand data and scenario expertise serving as its ace in the hole for AI office solutions.

Companies with long-standing experience in enterprise services indeed have a different perspective on the term "enterprise-level" compared to teams that originated purely in the consumer market. This is why, despite many claiming to offer AI office solutions, few have successfully landed at the enterprise level.

At the same time, enterprise-level AI office is a slow-moving business. The decision-making process, from initial contact to pilot programs, procurement decisions, and full-scale rollout, is lengthy, spanning months or even years. The outside world is accustomed to measuring everything with consumer internet metrics, such as weekly user growth.

But those metrics are irrelevant here. The team's ability to endure patience is far more critical than imagined.

In simple terms, the consumer market demands explosive growth, while the B2B market requires endurance—a strength closer to Alibaba's core competencies.

Of course, QwenWork also faces challenges. For instance, merging three products that served different user bases requires seamless integration of teams, experiences, data, and account systems. User perception of the agent's capabilities will also influence the degree and effectiveness of this integration.

The key issue is that advancing simultaneously on three fronts is a double-edged sword. On the positive side, it positions QwenWork ahead of competitors. However, if resources fall short, all three lines could stall. Alibaba still needs to fully penetrate one end to establish benchmark cases, making the stories for the other two ends more credible.

Another factor that cannot be ignored is that the window of opportunity for enterprise-level AI office does not belong solely to Alibaba. Traditional office software vendors like Kingsoft Office, as well as Tencent and ByteDance, have also entered the fray, while overseas giants like Microsoft are not sitting idle.

In this light, QwenWork's public beta feels more like a public declaration: Alibaba is doubling down on the enterprise-level AI office table. Whether Alibaba can win this race hinges on how deeply QwenWork integrates into users' workflows six months from now.

References:

Tang Chen's classmate, "WorkBuddy Turns Feishu into Doubao's 'Filling'"

Huxiu, "Alibaba's Two Secret Teams and Wukong's Restructuring"