Chinese tech giants escalate AI office race, shift to agent fine management

As competition in the AI industry shifts from general-purpose large language models to real-world application scenarios, AI-powered office tools have emerged as the central battlefield for internet giants in the next phase of competition. Leading companies are rapidly iterating products, integrating business systems, and heavily investing in the AI office赛道, engaging in a fierce battle over enterprise workflows, desktop interaction, and intelligent collaboration.

Tencent Leads with Mature Office Ecosystem

Tencent is leveraging its well-established office ecosystem to lead the market. The company has launched WorkBuddy, an intelligent office assistant that fully integrates core products such as WeCom, Tencent Docs, and Tencent Meeting, creating a closed-loop office experience across all scenarios.

ByteDance Restructures to Focus on AI Office

ByteDance recently completed a major organizational restructuring to concentrate resources on AI office development. The company has deeply integrated the capabilities of Feishu, Doubao, and Volcano Engine to launch the Doubao Work Product Suite. Leveraging Feishu's comprehensive office scenario advantages, ByteDance's AI office products can deeply engage with enterprise business data, autonomously decompose complex workflows, and execute cross-scenario collaborative tasks. The focus is on addressing the digital office upgrade needs of mid-to-large enterprises, targeting the high-end enterprise services market.

Alibaba Integrates Products Under "Qianwen Office"

Alibaba has also completed a business integration, upgrading multiple intelligent tools—including QoderWork, MuleRun, and Wukong—into a unified Qianwen Office offering. Powered by the Tongyi Qianwen large model, the suite focuses on code assistance, office automation, and enterprise data intelligence analysis. Alibaba emphasizes technical implementation and industry adaptation, targeting small and medium-sized enterprises and industry-specific office scenarios.

Baidu Enters with Lightweight "Baidu Dazi"

Baidu has launched Baidu Dazi, a lightweight entry into the mass office market. By offering low-barrier intelligent assistance features, the company aims to capture C-end user mindshare.

Competition Shifts to Deeper Dimensions

As major players intensify their efforts, the competitive logic of the AI office赛道 has evolved. The era of simply layering shallow AI features onto existing products is over. The current core competition now revolves around three key dimensions:

• Underlying technical architecture
• Full-process scenario adaptation
• Enterprise ecosystem integration

Basic functions such as text polishing and spreadsheet organization are now considered standard. The key differentiators for product success are:

• The ability to adapt to enterprise-specific workflows
• Achieving end-to-end automation
• Ensuring data security and compliance

Insights from Baidu Dazi Product Architect

"Office agents are primarily judged by their effectiveness and ecosystem," said Li Jingqiu, product architect for Baidu Dazi, in a recent interview with Securities Times reporters.

Baidu launched Harness at the end of March this year, which ranked first on both the PinchBench agent leaderboard and the Deep Research leaderboard. Harness demonstrates advantages in tasks such as deep research, content creation, complex logical reasoning, computation, and code generation, enabling it to complete complex tasks end-to-end across various office scenarios.

Ecosystem Integration

On the ecosystem front, companies are working to connect internal business units. For example, Baidu Dazi has already integrated with Baidu Search, Baike, and Wangpan, covering scenarios such as no-code development, marketing insights, legal services, and healthcare. Internally, Baidu Dazi serves as a unified entry point, aggregating Baidu's years of accumulated ToC and ToB applications. Some applications can be accessed and scheduled in the form of Skills, representing a relatively lightweight connection method.

Li Jingqiu noted that recently, many applications have adopted a workbench model. For instance, when a user requests the development of a literary-style hotel website, the system schedules a workbench, generates the application on the right side, and supports real-time editing and publishing. After integrating with data analysis capabilities, users can view and edit analysis results—such as store revenue—on the right side without needing to write SQL.

"Baidu Dazi not only connects internal applications but also links external partners through an open platform, providing services for more vertical industries," Li added.

Industry Openness and Enterprise Adoption

Notably, Li Jingqiu mentioned that the office intelligence industry has become more open this year. The enterprise version of Baidu Dazi supports login via Feishu and DingTalk accounts, and also integrates mainstream ToB products such as WeCom, DingTalk, and Feishu. Third-party office applications can be listed through suites, Skills, or connectors, helping users access data and applications from other office products to better execute tasks.

SaaS vendors can open up their capabilities in an Agent-schedulable manner, allowing their products to be used by both humans and Agents. Agent vendors, in turn, can integrate high-quality capabilities based on customer needs. This mutual openness and integration naturally fosters collaboration.

Enterprise Attitudes Toward AI Office

As various AI office applications emerge, the attitudes and investment levels of enterprises toward these products have drawn significant attention. In the interview, Li Jingqiu stated that enterprises are more advanced than people imagine.

"In the past, companies would ask whether purchasing Dazi could save labor costs. Now, some companies have begun to conduct detailed statistics—for example, measuring the proportion of work done by humans versus Agents in a given task. Enterprises want to quantify the value of Agents before deciding on the resources and budget to allocate."

"Enterprise adoption of Agents has shifted from last year's 'try first, widely experiment' approach to this year's 'fine-tuned management,'" Li said. Managers now evaluate the respective value of humans and Agents:

• Tasks that follow standardized SOPs and can be decided by Agents have greater automation potential.
• Tasks requiring subjective judgment and creativity are better suited for Agents to assist human workers.

Ultimately, enterprises are focused on how humans and machines can collaborate effectively, and what role Agents can play in the workflow.