English text is being prepared.

On the battlefield of large AI models, if consumer-facing chatbots compete on daily active users and traffic, then enterprise-facing office agents compete on real "productivity folding."

Recently, Analysys released the ranking "September 2026 China Desktop AI-Native Office Agent Platform Traffic."

Data shows that Tencent's WorkBuddy leads by a landslide with 85.34 million monthly visits, nearly twice that of the second place. Close behind are Alibaba's Qianwen Office (46.55 million) and ByteDance's Doubao Work (30.39 million), as well as ByteDance's TRAE (27.12 million) and CodeBuddy (25.88 million).

As one netizen on X lamented: "In this AI showdown, it looks like Tencent's WorkBuddy has already won. Baidu has fallen directly behind."

On this ranking, which reflects real desktop interactions, we see fierce fighting among the three giants Alibaba, Tencent, and ByteDance. But what is extremely lamentable is that Baidu, the first in China to proclaim "All in AI," has its flagship "Baidu Dazi" ranked only eighth with 9.26 million interactions, opening up a landslide gap with the front-runner camp. In another QuestMobile multi-platform monthly active user ranking, although Baidu Dazi ranks second, on the desktop side—where "deep embedding into workflows" is most tested—Baidu has indeed failed to reproduce its dominance from the search era.

Office AI agents, regarded as the golden track closest to the commercialization of large models and with extremely high per-customer prices, why do they show such extreme fragmentation? What exactly did Tencent do right? And why has Baidu gradually lost its voice on this battlefield where it should have been strongest?

An early start and a late arrival: Tencent's "interception" and WorkBuddy's dimensional strike

We analyzed in depth in a previous article that ByteDance was the first giant to place its bets on the office Agent direction. From the full AI transformation of Feishu to the launch of the Coze platform, ByteDance was once the strongest combination for "letting AI do white-collar work." But the one that ultimately achieved scale effects and took the absolute top position was the latecomer Tencent WorkBuddy.

The core reason WorkBuddy was able to achieve a terrifying 85.34 million visits is that Tencent precisely hit the most real "thrill points" of office workers.

First, it has an extremely lightweight "tool attribute" and frictionless experience. Compared with ByteDance's heavy integration strategy of trying to stuff Agents into the entire enterprise collaboration network, WorkBuddy takes a single-point breakthrough route: no deployment required, available on PC and mini-programs across multiple platforms, and users only need to input natural language instructions to directly generate reports, format documents, and process data. It does not require the whole company to cooperate on digital transformation; you can download it yourself and start using it immediately. This "plug-and-play" experience greatly lowers the barrier to adoption.

Second, it has an extremely smart model strategy—"no self-indulgence." WorkBuddy does not force users to bind to Tencent's own Hunyuan model, but fully embraces mainstream domestic open-source large models including DeepSeek and Zhipu. This "non-picky" open ecosystem allows users to freely choose the most cost-effective model best suited to the current task, reducing user migration costs to zero. For white-collar workers, they do not care what parameter level your model is at; they only care whether you can produce a chart in one minute that does not need rework.

The core barrier is Tencent's social foundation and ecosystem openness. When WorkBuddy seamlessly connects to WeCom, WeChat, and Tencent Docs, it possesses terrifying traffic entrances and context acquisition capabilities. After white-collar workers finish discussing work in WeChat, they can casually summon WorkBuddy to handle tasks, and can even directly forward generated files to WeCom groups. More notably, WorkBuddy is not completely closed within the Tencent ecosystem; officially it also supports integration with DingTalk and Feishu. This strategy of "relying on a big tree but not being exclusive" further broadens its user base. This seamless experience based on social relationship chains is a dimensional strike that no standalone app can match.

ByteDance and Alibaba's sprint: heavy integration and the cloud ecosystem game

The second and third places on the ranking are Alibaba's Qianwen Office (46.55 million) and ByteDance's Doubao Work (30.39 million). The routes of these two giants are completely different from Tencent's.

Alibaba is throwing a heavy punch of "cloud + collaboration." Qianwen Office (integrating products such as QoderWork, Wukong, and MuleRun) is deeply integrated into the DingTalk ecosystem and has extremely high penetration in the B-end enterprise services market. For large state-owned enterprises and manufacturing companies that emphasize organizational management, process approval, and data security, the combination of "DingTalk + Qianwen" is the default option for digital transformation. Alibaba's cloud ecosystem, cultivated for many years, is its deepest moat.

ByteDance is playing a deep reconstruction of the "family bucket." ByteDance has integrated Feishu, TRAE (code generation), and Coze (Agent building) all into the Doubao system, attempting to create an extremely deep "organizational super brain." This asset-heavy route of "building a system," once it works, will have an extremely deep moat, because what it obtains is not only surface-level work needs, but also the enterprise's most core underlying secrets and business flows. But for now, it is still digesting the friction period brought by intense internal restructuring, and its traffic temporarily lags behind Tencent.

The lagging Baidu: without an "entrance," even a good model is an island

Looking back at the most lamentable Baidu.

English text is being prepared.

The fundamental reason is this: Baidu lacks a true "enterprise-level collaborative office entrance."

This is an extremely cold business reality. In office scenarios, large models are not for writing poetry; they are for embedding into workflows. Tencent has WeCom and Docs, Alibaba has DingTalk, and ByteDance has Feishu. And Baidu? The visibility and penetration of Baidu Ruliu simply cannot support a massive Agent ecosystem.

In the competition for desktop office agents, users will not go out of their way to download an app just to use a particular model. What they need is: "When I'm writing a document, the AI can conveniently do the formatting for me; when I'm replying to WeCom messages, the AI can help me summarize the meeting minutes."

Baidu suffered from "lacking a super entrance" throughout the mobile internet era, and in the AI era, this fatal weakness is still dragging it down. It can only sell APIs, sell computing power, or pursue large customized To G (government) contracts, yet on the office battlefield that truly belongs to white-collar daily life, is high-frequency, and offers a C-end experience, it has lost the initiative. This is not only a tragedy of technology, but also the inevitable result of strategic path dependence.

The endgame of office AI: the leap from Chatbot to Agent

This ranking reveals an extremely clear industry trend: competition among large models has officially moved from the "chatbot" era into the "office agent" era.

In the past, everyone compared who wrote better love poems, who had stronger logical reasoning, and who scored higher on benchmarks. Now, everyone is comparing whose Agent can truly take over the operating system's mouse and finish the work for users.

WorkBuddy's victory proves that, at this stage, "lightweight entry + open model ecosystem + reliance on a super app" is the optimal solution for winning over white-collar minds. White-collar workers do not care how many parameters the underlying model has; they only care whether you can help them save overtime hours.

Of course, the current ranking does not represent the endgame. If ByteDance's heavy integration gets through its painful transition period, it may unleash an extremely formidable deep-context capability; Alibaba's cloud ecosystem remains the first choice for large B-end clients. Competition in office Agents is a marathon, and the current leaders must constantly guard against latecomers overtaking them on the curve.

But in any case, Baidu's performance on this ranking sounds a warning for all large-model companies: on the To B battlefield, a model without a traffic entrance is like a pearl in an exquisite box—no matter how dazzling, no one will use that box to serve rice on the family dinner table.

When large models shed the halo of technological myth, what ultimately determines victory or defeat are still those most plain business common senses: who is closer to users, who can solve specific pain points, and who can build an ecosystem network that lets everyone hitch a ride. AI without scenarios is destined to remain only in PPTs.