July has been a pivotal month for the AI-driven office sector.

Early in the month, a split occurred within QTeam, with QoderWork and a dozen employees breaking away to join forces with Wukong (incubated by DingTalk) and MuleRun (an internal startup at Alibaba Cloud). On the night of the restructuring, some employees began the move to DingTalk's campus in Hangzhou Future Sci-Tech City, completing their office relocation within a week. Alibaba later confirmed that the three products would merge into a new enterprise productivity suite, built on the foundation of QoderWork, with no disruption to existing user rights. The new offering, under the leadership of Chen Yusen (who assumed the role of DingTalk's CEO in June), was later christened Qianwen Office.

Within the same month, ByteDance underwent a similar restructuring. On July 30, the Feishu product team merged with the Doubao product team. Most Feishu employees learned about these changes simultaneously with the public announcement. Subsequently, Xie Xin began reporting to Zhao Qi (head of Doubao), while sales, marketing, and customer service teams were reassigned to Volcano Engine. Feishu's existing products will continue to operate, but a once-cohesive business line, spanning product development, R&D, sales, and delivery, was now divided.

Tencent made the least drastic changes. On July 20, QClaw-related businesses and teams were reassigned to Cloud Product Division VI (home to WorkBuddy). QClaw remains operational, and as of the press deadline, Tencent has not publicly announced any changes to WeCom's reporting structure. Among the three companies, Tencent made the fewest adjustments to its legacy office organization, but its new office assistant, WorkBuddy, has surged ahead. Analysys data reveals that WorkBuddy received 20.97 million monthly visits on the PC side in June, outpacing the combined total of second-place TRAE IDE (domestic version) and third-place QoderWork.

Alibaba streamlined its racing products, ByteDance divided Feishu, and Tencent let WorkBuddy take the lead. All three companies are consolidating their office assistants but are not making identical organizational shifts. After careful consideration, they redistributed responsibilities in office operations: who manages models, controls user gateways, stores enterprise data, and oversees contracts, delivery, and payments.

Previously, product gateways, enterprise data, and commercialization were primarily managed within DingTalk, Feishu, and WeCom's respective organizations. Now, the main gateway for work tasks is starting to shift away from these office departments.

DingTalk, Feishu, and WeCom are often lumped together in collaborative office market analyses, but they have never been identical products.

DingTalk is fundamentally a management tool. When Wu Zhao initially developed DingTalk, he expanded its scope from instant messaging to include attendance tracking, approvals, logs, low-code applications, and industry-specific tools. Employees use it for clocking in, leave requests, and reimbursements; managers rely on it to oversee organizational structures and processes; developers build enterprise applications on its platform.

DingTalk is more than just a communication tool for employees; it increasingly stores enterprise organizational relationships, approval rules, and business processes. Whether an employee belongs to a department, what data they can access, or who must approve a purchase—these details, though less visible than chats and documents, determine whether software can truly integrate into enterprise operations.

Wu Zhao left and rejoined DingTalk twice during its development. He resumed leadership in 2025 but stepped down again after 15 months. Before DingTalk's leadership change, a former employee published a lengthy post on Alibaba's intranet, critiquing the positioning and management of the AI-native project ONE. On June 11, Alibaba announced Chen Yusen as the new leader, who inherited more than just a mature business line. DingTalk must redefine its role in the AI era: continue as a direct office gateway for users or step back to provide organizational, permission, and process support through agents.

Feishu addresses a different challenge. Initially built as a collaboration tool for ByteDance's rapidly expanding workforce, it gradually opened up to external enterprises. Feishu believes that low enterprise efficiency stems not just from slow approvals but also from information silos across departments and hierarchies. In 2021, Xie Xin declared at Feishu 5.0's launch that "advanced teams use Feishu first."

At the same event, Feishu introduced HR, contract, and approval products, aiming to connect messaging, documents, meetings, and enterprise management. This statement effectively defined a work methodology: faster information flow across documents, group chats, and meetings, enabling more people to understand decision-making contexts. Feishu sells not just software but also ByteDance's collaborative experience in using it.

This approach requires sustained product and R&D investment, with growth rates unlikely to match those of consumer internet products. It also demands an enterprise sales system entirely distinct from ByteDance's consumer business. In March 2024, Xie Xin announced organizational streamlining in an all-staff letter, citing Feishu's large team size, declining efficiency, and lack of focus. Concurrently, Feishu shifted its commercialization focus to mid-to-large enterprises and annual subscription revenues.

While DingTalk manages organizations and processes, and Feishu handles knowledge and collaboration, WeCom evolved naturally from WeChat. It serves both internal enterprise communication and connects with customers and consumers on WeChat. Enterprise employees can manage customer relationships, group chats, and service interactions within WeCom, while users remain on WeChat without needing another app. In August of the previous year, Huang Tieming (head of WeCom) revealed that enterprises serve over 750 million WeChat users daily via WeCom—a feature that competitors find hardest to replicate.

The three software products each dominate a segment of the work chain, having built product teams, sales systems, and pricing models around their respective gateways. Previously, employees had to proactively access these systems to check messages, attend meetings, fill out forms, and manually transfer information between interfaces. Humans always bridged the gap between different software.

Agents reverse this sequence. Users first state their objectives, and the system then decides which documents to read, which meetings to review, which forms to access, and which business systems to engage. The starting point of office work thus shifts from software interfaces to the tasks themselves.

Big Tech companies are redirecting resources toward work scenarios not only because consumer products face temporary profitability challenges. Work completion yields relatively clear outcomes; time saved and processes shortened are easily quantifiable for enterprise procurement decisions. Tencent reported in its Q1 2026 earnings that WorkBuddy has achieved rapid growth and healthy retention; in June, Tencent Cloud launched an enterprise version, integrating office tools, code repositories, enterprise knowledge, and agent development tools into a single product ecosystem.

However, they quickly discovered that simply integrating AI into legacy office products does not naturally create an agent. Feishu, DingTalk, and WeCom had already incorporated AI into meetings, documents, Q&A, and business processes. These features could summarize meetings or generate text but did not alter the original work sequence: employees still confirmed tasks in group chats before opening documents, spreadsheets, approvals, and business systems separately.

Office agents demand a different outcome. Users no longer just ask AI for summaries but request analyses of sales data, identification of anomalies, PPT creation, and distribution of results to project teams. The system must independently locate materials, break down steps, operate software, and verify results. The product goal shifts from assisting individual actions to completing entire tasks.

However, individual efficiency does not automatically translate to organizational efficiency. Individuals can directly grant agents access to local files and computer permissions, but enterprises cannot do the same. For agents to integrate into organizations, they must connect with identity management, permission systems, business contexts, and audit trails. This means models, search engines, browsers, local files, enterprise knowledge, and sales delivery can no longer operate independently. When dispersed across departments, no single team could be held accountable for a task's final outcome.

Parallel multi-team efforts initially aimed to explore directions. As those directions became clearer, redundant products, computing resources, and sales systems turned into costs. The multi-track experimentation phase for office agents has thus entered a consolidation stage, with the next step being to reassign accountability for final task delivery.

The three companies' office restructuring first became evident in personnel changes.

Chen Yusen is not a traditional enterprise software manager. At 22, he founded cybersecurity company Chaitin Tech, later acquired by Alibaba Cloud. Entering this agent cycle, he led MuleRun. In a February interview with LatePost, Chen proposed that future software might be "disposable": as development barriers lower, users could quickly encapsulate personal work methods and processes into agents, use them temporarily, and regenerate them for new tasks.

Four months later, this advocate for rapidly generative and disposable software began managing DingTalk. Alibaba did not position Qianwen Office as just a feature within DingTalk but placed the same leader in charge of both the legacy office platform and the new agent product. After Chen's appointment, the R&D teams for Wukong and MuleRun merged first, followed by QoderWork's integration. The three products offered distinct capabilities: QoderWork excelled in desktop, local file, and programming access; Wukong connected with DingTalk's organizational scenarios; MuleRun accumulated experience in cloud execution and long-running tasks. Alibaba publicly confirmed that QoderWork would serve as the integration foundation.

This consolidation happened rapidly. On March 17, Alibaba launched Wukong, requiring CLI transformations to enable agent access to approvals, schedules, communications, and documents. Just four and a half months later, Alibaba began reining in Wukong's independent product trajectory. According to Tencent Technology, Wukong will subsequently integrate into the unified Qianwen Office system. While independent products converge, the enterprise agent direction they represent has gained a higher organizational position.

Having the same leader resolves questions of hierarchy. Qianwen Office needs DingTalk's customer base, organizational relationships, and business processes; DingTalk needs an agent capable of reclaiming user gateways. Placing both under the same reporting line reduces cross-departmental coordination and holds one team accountable for both product and enterprise scenarios.

However, organizational precedence does not equate to product completion. DingTalk is now a business requiring stability, with mature clients, revenues, and service systems. According to LatePost, the initial integration was hasty. QoderWork, Wukong, and MuleRun had different foundations, user bases, and teams, requiring gradual migration of capabilities and accounts. The true challenge lies not in merging the three products but in reallocating teams and resources.

While Alibaba consolidated power, ByteDance split Feishu's authority.

Previously, Feishu operated as a complete business line, with Xie Xin overseeing both product development and sales, marketing, and customer service teams. Entering large enterprise clients previously required months or even a year to sell an entire organizational work methodology, encompassing software deployment, training, and process adjustments.

The advent of AI turned this completeness into an issue requiring restructuring. For Doubao to enter office scenarios, it must access Feishu's documents, meetings, group chats, and organizational permissions. For Feishu to develop agents, it needs models, browser operations, and local system access. Volcano Engine's enterprise clients simultaneously demand cloud services, models, and office products.

The three teams now serve the same user base with overlapping capabilities. As early as September 2025, to strengthen office collaboration between Doubao and Feishu, a Feishu product leader began reporting dotted-line to Zhao Qi. In June 2026, Doubao publicly acknowledged collaborative efforts with Feishu in enterprise office scenarios, promising deeper integration.

The new organizational adjustment formalized this relationship: Doubao handles user tasks, Feishu provides enterprise context and collaboration, and Volcano Engine manages sales, cloud services, and model delivery. Feishu's official description of Doubao Enterprise Edition reflects this structure: employees initiate tasks through Doubao or Feishu; Doubao autonomously breaks down steps and accesses cloud computers, browsers, and Feishu tools; generated documents and spreadsheets return to Feishu, where they can be edited, commented on, and shared using existing organizational permissions.

Tencent adopted a distinct strategy. Initially, it kept WeCom unchanged, opting to allow WorkBuddy to develop autonomously outside the existing office ecosystem. According to LatePost, in January, following the launch of Claude Cowork, several members of the CodeBuddy team developed an early version of WorkBuddy over a single weekend. They repurposed CodeBuddy's infrastructure, substituting developer-specific jargon with terminology more comprehensible to general office workers. The product featured a streamlined single dialog box, enabling users to specify their objectives without needing to understand MCPs, Skills, or Agents.

Unburdened by the constraints of legacy products, WorkBuddy initially focused on addressing immediate personal tasks: reading files, generating reports, analyzing data, creating PowerPoint presentations, and operating computers. After establishing a foothold in personal use cases, Tencent began integrating its capabilities. WorkBuddy now interfaces with Tencent Docs, Meetings, Email, enterprise knowledge bases, and segments of the WeChat ecosystem. Tencent Docs can be directly read and edited, with the results seamlessly written back.

Tencent has officially positioned WorkBuddy as an AI workplace agent capable of autonomously breaking down and executing multi-step tasks. The enterprise version consolidates office tools, code repositories, and managed agents into a unified platform. Pricing remains seat-based but includes a per-seat allocation of Credits, with additional usage available for purchase. This approach combines traditional SaaS subscription models with token-based consumption into a single pricing structure.

Tencent initially achieved product-level integration while maintaining parallel organizational structures. WorkBuddy excels at personal work tasks, WeCom manages organizational and customer relationships, Tencent Meetings and Docs store office content, and Hunyuan provides the foundational models. These components operate within different organizational units without forming a centralized power structure akin to Alibaba's. This parallelism may alleviate historical burdens on new products but introduces complexity in the next stages of development.

Once WorkBuddy is deployed within a company, several questions arise: How should it collaborate with WeCom? How should WeCom's own Agent be positioned? Who will be responsible for the model, customer contracts, and Token revenue?

In essence, the three organizational structures are all competing for the same prize: who will be the first to respond to an employee's initial task request.

Large corporations are willing to reconfigure their office organizational structures, not because DingTalk, Feishu, and WeCom have lost their relevance. On the contrary, these platforms retain what Agents require most to penetrate enterprises: employee identities, organizational relationships, business data, and permissions and audit rules.

An Agent must not only comprehend goals, manipulate documents, and interact with business systems but also ascertain which department a user belongs to, what data they can access, and whether they possess the authority to send messages and modify processes. Models and desktop Agents primarily address the former aspects, while DingTalk, Feishu, and WeCom accumulate the latter. This is why, even after some diminishment of their gateway value, these previous-generation office tools remain irreplaceable.

Previous-generation office software centered on instant messaging and interfaces. Employees would typically initiate the product and then navigate to documents, meetings, attendance, and approvals via menus. Part of the product's value derived from keeping employees within the same interface for extended periods. Agents invert this sequence: users first articulate their goals, and the system then determines which software to invoke. Documents store content, meetings provide records, WeCom offers relationship chains, DingTalk manages processes, and Feishu supplies knowledge. However, these tools no longer necessarily serve as the starting point for work.

Communication will not vanish but may shift later in the workflow. Previously, work often commenced with meetings, discussions, and assignments before employees executed tasks. Now, Agents can first generate plans and preliminary results, around which humans then conduct reviews. Instant messaging remains crucial but may no longer constitute the first step in the work process.

The value of enterprise software is thus bifurcated into two layers: the front end belongs to Agents capable of understanding goals, planning tasks, and delivering results, while the back end comprises data, permissions, audits, and business rules. The emphasis on UI and menus may diminish, but organizational context will become more valuable. Feishu highlights in its Doubao Enterprise Edition that Agents inherit employees' existing permissions, with enterprise data isolated from personal data. WorkBuddy Enterprise Edition offers shared cloud, dedicated VPC, and private deployment options, enabling enterprises to manage Credits, logs, and usage scope.

Billing methods are also evolving in tandem with this division. Traditional collaborative office tools primarily charged by seat and version, but Agents continuously invoke models, cloud environments, and external tools, with costs depending not only on the number of employees but also on task frequency, duration, and complexity.

WorkBuddy Enterprise Edition still prices by seat but includes 2,000 Credits per seat. The SaaS Enterprise Edition and Private Cloud Enterprise Edition cost 198 yuan and 316 yuan per person per month, respectively. If usage falls short, additional Credits packages must be procured. Enterprises still purchase software based on headcount, but Agent usage beyond the basic quota incurs separate fees.

Once customers commence payments, new questions emerge, such as who signs the contracts, which department receives the revenue, and who bears the inference and delivery costs.

When Doubao initiates tasks, invokes Feishu Documents, and utilizes Volcano Engine models, are customers paying for the Agent gateway, collaboration software, or computing power? When Qianwen Office penetrates enterprises via DingTalk, how is the new revenue allocated from DingTalk's existing seat and service revenue? When WorkBuddy invokes WeCom, Tencent Meeting, and Hunyuan, who signs the contracts, bears the inference costs, and receives the revenue?

This indicates that competition among office Agents will not be confined to model capabilities after they enter enterprises. It also hinges on whether a company can simultaneously integrate four layers: models, task gateways, enterprise data, and sales delivery. Thus far, none of the three companies have fully connected all four layers.

Alibaba has centralized power under Chen Yusen, reducing coordination layers between old and new products but necessitating resource allocation between DingTalk's stable revenue and Qianwen Office's rapid trial-and-error. ByteDance has positioned Feishu between Doubao and Volcano Engine but still needs to demonstrate that three interconnected pathways can operate more efficiently than a standalone Feishu. WorkBuddy can commence as a personal productivity tool, but in complex enterprises, organizational permissions, customer relationships, and industry-specific delivery will become increasingly critical. It currently leads in traffic but has not publicly achieved organizational alignment across models, Agents, and WeCom.

Traffic does not equate to stable task completion, personal user activity does not equate to enterprise willingness to share core data, and seat purchases do not equate to revenue covering model, cloud, and delivery costs. Organizational adjustments merely clarify responsibility. The next phase requires observing whether a real task can be completed, how often users must intervene, which permissions enterprises grant, whether they renew after six months, and whether revenue per seat covers model invocation, cloud operations, and delivery service costs.

WorkBuddy has secured first-stage advantages in traffic and product rhythm, but this does not ensure ultimate success in the enterprise market. However, this advantage has already prompted the other two companies to reposition their forces.

In the past, DingTalk, Feishu, and WeCom represented each company's response to office productivity. Wu Zhao digitized enterprise organizations and processes, Xie Xin once endeavored to persuade enterprises to adopt a new collaboration method through "advanced teams," and Huang Tieming connected internal employees with external WeChat clients. They not only controlled their products but also influenced how office work was defined.

Agents offer everyone a chance to start anew. Employees no longer need to embrace a management philosophy before learning complex software. As long as the results are satisfactory, an Agent can transition from personal tasks into enterprises, gradually integrating with existing organizational, knowledge, and business systems.

The old fortresses are not devoid of value. What is being redistributed are the front-end gateways, while employee identities, organizational permissions, enterprise knowledge, and customer relationships endure. Alibaba has entrusted these assets to Chen Yusen for recombination, ByteDance has disassembled Feishu into Doubao and Volcano Engine, and Tencent has allowed WorkBuddy to operate outside WeCom before aligning the two.

The product competition is still far from over, but organizational judgments have already been rendered. Even if the previous-generation office tools do not vanish, they will no longer automatically command center stage.

*The featured image and illustrations in the text are sourced from the internet.