Author: Ren Tianqin

Editor: Wang Yimo

Two Technological Paths Initiate Long-Term Competition

Internet giants are actively deploying desktop agents, branching into two distinct development paths: self-developed foundations and multi-model aggregation. A contest for the next-generation office access point has officially shifted from product launches to competing for real users.

Offline Competition for User Mindshare

Over the past few months, leading domestic internet companies have intensively advanced their desktop office agent product layout (product layout translates to 'product layout ' but in context should be 'product layout') and product line integration, officially igniting a fierce battle for the 'next-generation enterprise office access point.'

Amid this competitive landscape, major players have unveiled their trump cards: Alibaba has deeply integrated three products—QoderWork, Wukong, and MuleRun—to launch 'Qianwen Office,' aiming to leverage DingTalk's vast B-end ecosystem to create an AI productivity platform for enterprise organizations.

Tencent has concentrated its resources to promote WorkBuddy, a product that deeply integrates WeChat and Enterprise WeChat. Leveraging the viral advantages of its social ecosystem and an extremely low barrier to trial, WorkBuddy has achieved a dominant lead in monthly visits.

Baidu has launched the desktop-level AI agent Baidu Dazi DuMate, which can access Baidu ecosystem capabilities such as cloud storage and document libraries through skill components, integrating existing product resources.

ByteDance has adopted a dual-track approach, advancing Doubao Office, which is deeply integrated with Feishu, while also promoting TRAE Work's expansion from developer-exclusive to universal office scenarios for all employees.

With multiple products launching in quick succession, all targeting high-frequency office needs such as document writing, data analysis, and plan organization, the Track ( Track translates to ' Track ' but in context should be 'sector') homogenization of competition continues to intensify.

However, the underlying logic of competition is undergoing a profound transformation. AI office tools are evolving from 'answering questions' as auxiliary tools to independent executive agents that 'work for users.'

Thus, the current competitive focus has shifted from merely 'whose model parameters are stronger' to 'whose task success rate is higher and unit cost is lower,' as well as who can become the first interactive interface users encounter when they turn on their computers.

In the future, the ultimate outcome of this office agent competition is more likely to evolve into tiered competition. Whoever can first deeply embed AI into enterprise workflows, enabling users to develop irreplaceable 'muscle memory,' will be poised to defend and reshape the distribution landscape of enterprise software.

According to Analysys data, by June 2026, the combined monthly visits of 17 mainstream desktop AI office agents in China exceeded 60 million. Among them, Tencent's WorkBuddy led with 20.97 million monthly visits, followed by ByteDance's TRAE IDE domestic version with 12.79 million.

Big companies are expanding their marketing beyond online channels and actively deploying offline community advertising, indicating a shift in the industry's competitive focus.

A Dual-Pronged Approach to Secure Overall Success

Behind the fierce market competition lies a bold strategic adjustment within the giants, with ByteDance's restructuring being highly representative. In late July, ByteDance completed the integration of Doubao, Feishu, and Volcano Engine businesses.

The previously independently operated Feishu team was split, with the Feishu product team merged into the Doubao system and the reporting line of its person in charge ( person in charge translates to 'person in charge') adjusted; the Feishu commercialization GTM team was integrated with Volcano Engine to form a ToB service organization named 'Creativity Service Platform.'

Despite Feishu's steady business growth, with most new clients also purchasing AI-related services, ByteDance's push for organizational restructuring underscores the group's commitment to AI office solutions.

After the restructuring, the Doubao Enterprise Edition, built on Feishu's document, meeting, and group collaboration capabilities, has initiated targeted internal testing among existing Feishu clients, integrating a complete enterprise-end workflow.

While the outside world's attention is primarily focused on Doubao Office, it is TRAE Work that has first engaged in offline competition with WorkBuddy, highlighting ByteDance's typical dual-track strategy.

Doubao Office is rooted in the B-end enterprise collaboration ecosystem, leveraging Feishu's resources to serve organizational clients; TRAE Work extends beyond programming tools to target individual professionals.

By advancing both product lines simultaneously, ByteDance hedges risks internally and comprehensively covers the C-end and B-end markets externally, attempting to find the optimal commercialization path in this yet-to-be-defined sector.

No Standard Answer Yet

Within the sector, product goals are converging, but underlying technological paths are clearly diverging.

Qianwen Office and Doubao Office have chosen the self-developed large model route, relying on their proprietary foundations for continuous iteration, with model capabilities directly setting the upper limit for agent task processing. TRAE Work and WorkBuddy advocate for a multi-model aggregation approach, integrating multiple large models into a unified workbench.

Both paths present their own challenges, and the industry has yet to find a standard answer.

The self-developed route requires sustained heavy investment in computational resources, covering both model training and online inference costs; the core challenge of the aggregation route lies in intelligent scheduling strategies—how to automatically match the optimal model for different tasks while balancing inference costs and output effectiveness remains an engineering challenge the industry urgently needs to overcome.

The industry's growth potential is vast. IDC predicts that by 2026, the Chinese enterprise-level AI agent market will reach RMB 44.9 billion, indicating significant long-term growth prospects.

According to iiMedia Research, the usage rate of AI assistants among white-collar workers in China has exceeded 57%, with penetration rates continuing to rise in scenarios such as document editing and data analysis. Given the abundance of repetitive tasks in office settings, intelligent agents offer clear value in cost reduction and efficiency enhancement, ensuring the sector's long-term development potential.

It is important to note that the industry is still in its early stages, with user payment habits not yet fully cultivated and more enterprise pilots than large-scale implementations. Data shows that while over 60% of enterprises have incorporated AI agents into their medium- to long-term digital strategies, only 17% have achieved company-wide Large scale implementation ( Large scale implementation translates to 'scaled implementation').

Model hallucinations, enterprise data security, and sustained high invocation costs are all significant factors restricting commercialization progress.

In essence, the competition for office agents is a battle for control over the next-generation computer desktop.

Regardless of the technological path chosen, only by continuously addressing real office pain points, integrating complete workflows, and establishing a sustainable business model can companies stand firm in this prolonged AI productivity competition.