Tencent, ByteDance, Alibaba vie for AI office market, consolidate products into single-entry platforms

No one knows exactly where this path leads.

By Xu Junhao | Haike Finance

In the surging wave of AI, the office productivity scenario has become the primary battleground for leading players.

The intense competition under the spotlight is clear evidence. After completing organizational restructuring, ByteDance, Tencent, and Alibaba have all concentrated their forces. On September 15, ByteDance CEO Liang Rubo made a rare appearance at the Feishu Future Unlimited Conference and Doubao Work Launch Event, announcing the formal integration of Doubao, Feishu, and Volcano Engine's capabilities, increasing investment in the enterprise-level AI office market, and launching new products such as Doubao Work Partner.

Tencent's WorkBuddy has been moving rapidly. WorkBuddy launched its open platform on September 2, initially bringing in over 100 ecosystem partners; on September 8, WorkBuddy cross-industry partnered with the game Peacekeeper Elite; on September 16, it launched the WorkBuddy Travel Tool Creation Contest.

Alibaba's Qianwen Office released its first-month performance report on September 4. Official data shows that since its launch, Qianwen Office has completed 120 version updates, iterating an average of 4 times per day; registered users have exceeded 30 million, with over half being enterprise users.

Other players are also carving out their positions in various ways. Recently, Baidu announced it would build Baidu Dazi into a fully open ecosystem platform, simultaneously opening for入驻 and launching partner recruitment, opening up four major capabilities—connectors, skills, suites, and hardware—for capability providers, and pledging never to charge commissions from partners. On September 14, Kingsoft Office officially obtained the AI agent identity code node, becoming the first domestic manufacturer in the AI office field to hold such a node.

Overseas leaders are even more eye-catching. Microsoft Copilot, leveraging Office's long-standing积淀, holds a first-mover advantage; Google's Workspace is accelerating its catch-up; OpenAI launched ChatGPT Work in July 2026; Anthropic merged Claude Cowork and Claude Chat into Claude on September 17.

According to the 2026 China AI Office Agent Industry Development White Paper recently released by research firm iiMedia, the global AI agent market is undergoing a critical transition from technical validation to large-scale implementation. AI office agent products will emerge intensively, becoming the backbone driving the steady growth of the entire agent industry. The global AI agent market size was $37.2 billion (approximately RMB 249.4 billion) in 2025, is expected to reach $69.8 billion (approximately RMB 468.1 billion) in 2026, and could reach $312.2 billion (approximately RMB 2.0937 trillion) by 2030. The domestic AI agent market size was RMB 80.4 billion in 2025, is expected to reach RMB 155.8 billion in 2026, and could reach RMB 696.8 billion by 2030.

As overseas model capabilities continue to突破, AI office, as one of the benefiting application branches, rises accordingly. Driven by FOMO (Fear of Missing Out), domestic tech giants are pouring in resources and全力投入, making the落地 speed of AI office far exceed other tracks. But the frenzy cannot掩盖 uncertainty. In the current era of rapid AI development, there are still many variables.

From Lobster to Office

The growth trend in the AI office field is evident to all.

According to data from research firm QuestMobile, in July 2026, the overall MAU (Monthly Active Users) of the domestic AI efficiency office track reached 102 million, a year-on-year increase of 8.2%. Among them, the native APP端 MAU of the office track was 7.926 million, up 261.1% year-on-year, and the PC client MAU was 10.301 million, up 340.6% year-on-year. Calculated by MAU, the top four office agents were Tencent WorkBuddy (6.582 million), Tencent QClaw (2.259 million), ByteDance TRAE Work (1.904 million), and Alibaba QoderWork (230,000).

It should be noted that this field has changed extremely rapidly recently. In July-August 2026, Tencent, ByteDance, and Alibaba successively ended multi-track horse racing, consolidating their previously分散 product matrices into a single entry point. Tencent merged QClaw into WorkBuddy on July 20; ByteDance merged the Feishu product team into Doubao on July 30, with Feishu CEO Xie Xin reporting to Doubao product head Zhao Qi. Subsequently, on August 24, the TRAE and Kousi teams were merged into Doubao, and on August 25, the independent brand "Doubao Work" was launched. Alibaba integrated QoderWork, Wukong, and MuleRun into "Qianwen Office" in July, managed by DingTalk CEO Chen Yusen, and Qianwen Office was officially released on August 3.

The industry generally categorizes Baidu Dazi and Kingsoft Lingxi into the domestic second tier, with their current visibility lower than that of Tencent, Alibaba, and ByteDance. Baidu Dazi was launched in March 2026 and completed upgrades for personal, enterprise, and workbench versions in August; Kingsoft Office released Lingxi Professional Edition and WPS Comate in July 2026.

According to Haike Finance's observations, in terms of product form, the各家 are largely similar. The basic architecture uses tasks or conversations as the main interaction window, supplemented by functions such as scheduled tasks, connectors, experts, skills, and IM channels.

A "connector" refers to a standardized interface between the agent and the enterprise's existing software systems. For example, connecting to office applications like DingTalk or Feishu allows the agent to access the application and read/write content within it. An "IM channel" connects to instant messaging tools, enabling the agent to connect to applications with IM functions like DingTalk, Feishu, WeCom, and WeChat to receive or reply to user messages.

Currently, WorkBuddy, Qianwen Office, and Doubao Work can connect to a wide range of applications, including Feishu, DingTalk, WeCom, as well as Office, Kingsoft Documents, etc. Baidu Dazi and Kingsoft Lingxi have far fewer connectable applications. In terms of expert and skill configuration, the leading players also have substantial reserves. The aforementioned Doubao Work Partner is essentially a pre-packaged capability module within the agent, ready for users to use without needing to configure from scratch.

The origin of the agent form can be traced back to the "Lobster Craze" in early 2026. "Lobster" refers to the open-source personal AI assistant OpenClaw, the original form of the agent. After local deployment ("raising the lobster"), the lobster could operate the computer on behalf of humans to complete tasks. However, OpenClaw had a high barrier to entry; "raising the lobster" required a certain level of technical skill, and the agent's Harness (engineering runtime system) was also immature.

Subsequently, various lobster-like agents emerged: Tencent had QClaw, Alibaba had JVS Claw, ByteDance had ArkClaw, and Baidu had DuClaw. The standout was WorkBuddy, launched in March 2026. WorkBuddy eliminated complex code configuration steps, supporting single-sentence natural language commands to initiate tasks, with the model automatically decomposing and planning and directly outputting complete results. More importantly, WorkBuddy is based on WeCom. After downloading and installing, users connect to WeCom and can remotely assign tasks to the desktop client via mobile phone, completing configuration in as fast as 1 minute. This is extremely user-friendly for ordinary users.

The AI office form originally popularized by Microsoft via Office was Copilot, which helps users draft, generate formulas, and summarize emails in a sidebar—still an辅助 form. However, the vast majority of office work is not single-step Q&A, but multi-step, cross-tool tasks with clear deliverables. These tasks have clear goals, decomposable steps, callable digital tools, and verifiable results, making them ideal scenarios for autonomous agent execution.

Executing a long-range task often requires dozens or even hundreds of model calls. If the cost of a single call is too high, large-scale implementation is根本 impossible. Now, with the rapid development of large models, model capabilities are quickly improving, and costs are decreasing. These conditions combined have共同 contributed to the prosperity of office agents.

Players Leverage Their Strengths

Compared to general-purpose agents, the core advantage of office agents lies not in how powerful the model is, but in being convenient and easy to operate, while being able to directly connect with the enterprise's existing office software and embed into the workflows users are already using.

For this reason, domestic players have taken截然不同的 paths based on their own genes.

Although agent connectors can connect to other players' products, the native capabilities embedded within each company's original products differ, and this is the key to building unique advantages. Tencent's strength is the social入口 formed by WeChat and WeCom. WeChat covers C-end reach, while WeCom handles B-end execution. WorkBuddy allows AI capabilities to渗透 into workflows along existing social relationship chains and enterprise organizational structures. WorkBuddy also adopts a model aggregation platform format, integrating not only Tencent's Hunyuan model but also external models like DeepSeek, GLM, and Kimi, allowing users to switch based on task needs.

The paths of Qianwen Office and Doubao Work are more similar. Qianwen Office briefly上线 GLM and DeepSeek external models but quickly returned to the Qwen model. Doubao Work has always used only the Doubao model. Both have accumulated considerable user bases on the C-end and then further developed based on the strengths of their original office applications. According to QuestMobile data, in June 2026, DingTalk's MAU was 171 million, Qianwen's MAU was 167 million; Feishu's MAU was 30.19 million, and Doubao's MAU was 382 million.

Qianwen Office possesses native capabilities such as reading DingTalk group chat messages, generating documents and spreadsheets, scheduling meetings, managing attendance and approvals, and reading/writing knowledge bases and emails. It transforms information scattered across messages, documents, approvals, and emails into organizational context that the agent can understand and call upon.

Doubao Work shares the same account and permission system with Feishu. Within authorized scope, it can call upon information from group chats, documents, meetings, multidimensional tables, and enterprise business systems. After employees log in with their Feishu accounts, the agent inherits their existing permissions. There is no need for each employee to re-upload materials or create a separate account system. All organizational context left by meeting decisions, information permissions, task assignments, and responsibility attribution can be directly called upon by the agent.

Baidu Dazi and Kingsoft Lingxi are betting on document assets. Baidu relies on its existing stock from search and Baidu Wenku; Kingsoft Office relies on its long-term accumulation in document editing, collaboration, and archiving. The reason Tencent, Alibaba, and ByteDance are stronger is that they can provide the context that agent RAG solutions most lack. RAG, or Retrieval-Augmented Generation, is one of the mainstream solutions for large model applications. After a user asks a question, the system first retrieves relevant document fragments from the knowledge base and then hands them to the model to generate an answer. Traditional RAG or document-based RAG can retrieve content but lacks understanding of key office elements like organization, collaboration, hierarchy, and processes. WeCom, DingTalk, and Feishu are inherently where organizational collaboration occurs. The agent no longer reads isolated document fragments but the entire context of organizational operation.

Currently, these products all adopt a model of free quotas plus积分 subscription, calculating积分 consumption in real-time based on the tasks completed by the agent. For example, WorkBuddy's free version provides 500积分 per month, and users can also earn积分 for free by completing tasks. The consumption for completing a single task like making a PPT or collecting information ranges from a few to dozens of积分. For paid subscriptions, different积分 package versions have monthly fees ranging from RMB 70 to RMB 999.

The commercialization of leading players is now showing initial results. During Tencent's Q2 2026 earnings call in August 2026, Tencent CSO James Mitchell mentioned that the gross margin of WorkBuddy paying users and the gross margin of Tencent's model services are already comparable to Tencent Cloud's overall gross margin. WorkBuddy's overall gross margin is relatively lower because it includes some free users. Tencent subsidizes these users to drive market share growth, while the paid portion has already generated quite good gross margins.

The Endgame May Not Be Far

In the AI office field, internet giants have strong capital, holding ecological, entry, and organizational execution advantages. These advantages are real.

A research report released by CITIC Securities on August 31, 2026, mentioned that the potential of the AI office scenario is enormous. The global potential user base could reach 1 billion white-collar workers, and in an optimistic scenario, it could create a trillion-dollar market space. Internet vendors have突出 competitive advantages in the AI office track, mainly reflected in ecological advantages, strong product capabilities, and the ability to route models at a low threshold. That is, the platform can automatically distribute different user requests to the most suitable large model behind the scenes. Users do not need to evaluate, select, and对接 models one by one. The access threshold is low, and a balance between效果 and cost can be achieved.

In the mobile internet era, where leading players' MAU often exceeded 100 million, the AI office's tens of millions of MAU may seem small in scale, but the competition among players is intensifying. Many users have lamented on social media that advertisements for various office agents are everywhere—at airports, in office buildings, and even on subways.

It's important to note that the battle for C-end user acquisition in large models has already concluded. Around the 2026 Spring Festival, leading players in the large model track全力 marketing, promoting their respective products through red packets, subsidies, and other forms, and indeed achieved some results. However, the larger the C-end user scale, the higher the computing cost. The叠加 of user acquisition costs and computing expenditure only makes the investment heavier and harder to bear in the long run. Consequently, vendors have had to change strategies. For example, Doubao launched a paid model in June 2026.

There is also fear. Under the strong冲击力和颠覆性 of AI, internet giants' concerns about falling behind or even being abandoned by the times are growing daily. Official data shows that in April 2026, Anthropic's Claude Code had only about 1.4 million weekly active users, but its同期 ARR (Annual Recurring Revenue) exceeded $10 billion (approximately RMB 67 billion). This means that the high-value economic入口 brought by AI may not be built on high user scale. If AI can redefine high-output fields like coding, the returns can multiply. Once agents redefine professional work in the office domain, what leading players lose is not just user scale but the qualification to compete in the entire track.

However, the war between DingTalk, Feishu, and WeCom themselves continues, and AI office agents have not yet拉开 a significant gap.

Many bloggers have already compared various AI office agents, such as "Jiguo AI Review Room" and "Peter Pan AI." They conducted实测 on items like meeting summaries, PPT generation, spreadsheet cleaning, scheduled tasks, and multi-device collaboration. The conclusion is that each player has its strengths, but there is still a gap from true intelligence. For example, when collecting images and data to make a PPT, some agents' layouts are unsuitable for actual PPT presentation, some paste the wrong product images, and some have data that doesn't match the product. Once the task chain lengthens, involving multi-step reasoning, cross-system operations, or dynamic adjustments, problems easily arise.

Domestic and overseas AI office are separated by compliance barriers. Domestic office agents' operations are deeply embedded in collaborative office platforms like DingTalk, Feishu, and WeCom,深入 enterprise processes. Overseas, giants including Microsoft Copilot and Google Workspace are competing for market share. However, OpenAI and Anthropic,公认 as the most technologically领先 AI players, have not separately established office product organizational structures. "Work" is just a mode and entry point within their respective product matrices. Anthropic's Claude Cowork product took only about 10 days from conception to launch, being a derivative incubated by the Claude Code team using Code's own capabilities.

After Anthropic merged Claude Cowork and Claude Chat into Claude, users no longer need to pre-judge whether to assign a task to chat or to a specific agent. Claude automatically calls upon capabilities like search, file analysis, code execution, spreadsheets, and PPT based on task complexity.

OpenAI is also pursuing a similar path. The Codex independent product line was terminated in April 2026, with its capabilities integrated into ChatGPT.

When OpenAI released GPT-6 Astra on September 3, OpenAI co-founder and President Greg Brockman said, "Welcome to the AGI era." AGI, or Artificial General Intelligence, refers to AI that can understand the world like a human, reason autonomously, and flexibly handle complex cross-domain tasks. GPT-6 Astra can already simulate humans directly reading screen pixels, meaning it operates like a human, without needing to rely on software vendors' API interfaces like agents do.

If AGI is truly achieved, theoretically, AGI could handle everything for humans directly, including work. At that point, the office scenario would be merely a subset covered by AI capabilities, and the necessity of the office agent as an intermediate layer might be minimal.

Under the wave, players do not intend to stop and论证. They simply choose to埋头 continue moving forward.

Disclaimer: The above content (including images and videos if any) is uploaded and published by a user of the NetEase Hao self-media platform. This platform only provides information storage services.