The Second Round of the AI Office Battle Begins

11/16 2026 356

Source | Bohu Finance (bohuFN)

Author | Qin Sanqian

Recently, major domestic tech companies have successively shifted their focus from personal intelligence to organizational intelligence.

In mid-August, ByteDance renamed Feishu Aily to Doubao Work Partner, subsequently launched proactive work capabilities, and worked with Feishu 8.0 to rebuild the underlying system. Tencent then launched the WorkBuddy Enterprise Edition and the WeCom joint suite, using Buddy One to connect a fragmented office ecosystem. Robin Li also proposed the concept of the "intelligent organization" at Baidu World 2026 and launched the Token Institute, which measures the authenticity, quality, and credibility of enterprise model services.

Various signs indicate that the focus of AI office is shifting from serving individuals to serving teams and organizations.

Capabilities such as understanding context, invoking tools, and delivering results—previously regarded as belonging more to the realm of personal productivity—have now entered the organization's metrics as well.

Among the many product paths, Alibaba's direction appears somewhat distinctive. On the afternoon of November 9, Alibaba Qianwen Office released version 2.0, introducing capabilities such as multi-person collaboration, human-machine co-creation, a material space, automatic tasks, and three-terminal interconnection.

Over the past decade, daily collaboration in Chinese enterprises has relied heavily on IM group chats. Creating a group is almost the starting point for all cross-department business, and group bots are also a common form of early office AI entering IM.

But this time, Qianwen Office is trying to shift the focus so that organizational collaboration revolves around Agents. It supports users in directly creating collaboration spaces within the client and then linking multiple DingTalk groups, online documents, AI tables, and business materials into them.

It can be seen that when AI enters real organizations, it does not quickly erase the boundaries of existing office software. Instead, it must first rely on the assets accumulated by Feishu, WeCom, and DingTalk over the past decade, and rearrange the positions of Agents and IM within the existing organizational order.

01 How Should AI in Organizations Run?

"Context is all you need." More than a month ago, Chen Yusen, CEO of Qianwen Office, offered his core judgment on enterprise-level Agents at the Yunqi Conference.

To deliver results at the organizational level, context is the fundamental condition determining the usability of Agents.

But in companies that actually operate, business context is never fed to AI out of thin air. In addition to various business systems established in the SaaS era such as ERP, CRM, and email, the IM that pops up and scrolls day after day is also one of the important sources of organizational context.

For enterprises, which IM to use often involves long-term organizational investment. The address book structure, approval flows, historical document permissions, and even the communication methods employees are accustomed to are all embedded in the platform.

Compared with the possibility of piloting point tools, the cost for enterprises to directly migrate the underlying IM is much higher. This means that when office Agents provide organizational intelligence services, following the platform endowments accumulated in their respective histories is the path of least resistance.

At present, ByteDance, Tencent, and Alibaba are all reconstructing the relationship between Agents and IM, deploying independent workbenches or client entrances while also retaining native IM access capabilities.

ByteDance follows Feishu's density of knowledge and documents, making Agents a type of organizational member within Feishu.

Since its birth, Feishu has emphasized information transparency, cloud document co-creation, and multi-dimensional table accumulation, gathering a large number of knowledge-intensive teams. Feishu 8.0 reconstructed its data, tools, and permission systems around Agents.

Therefore, its office Agent product Doubao Work Partner, in addition to having an independent PC client and Doubao entrance, is also specifically embedded in Feishu group chats and document scenarios. Its core value lies in invoking existing knowledge, comment records, meeting minutes, and schedules within employees' permission scope to jointly advance tasks with the team.

Looking at WorkBuddy, as one of the earliest office Agents to emerge, its implementation path relies on a multi-application ecosystem and connectivity advantages, using an independent workbench to connect information scattered everywhere.

Since its launch, WorkBuddy has been positioned as an independent desktop workbench, supporting access to communication tools such as WeCom, Feishu, and DingTalk, and commercially providing two benefit plans: the Buddy One agent suite and the WeCom joint suite. The typical problem it aims to solve is cross-application invocation, gathering materials originally scattered across different software windows and customer communications to complete end-to-end delivery.

Alibaba, relying on DingTalk, cuts into hierarchical processes, permissions, and heavy business systems.

Over the past decade, DingTalk's deepest moat has been moving Chinese enterprises' organizational structures, OA approvals, attendance to-dos, and business flows online. As of early 2026, third-party monitoring shows that its MAU has stabilized at around 200 million.

The demand for hierarchical management has always been engraved in DingTalk's product DNA. From the early Ding with strong reminders to last year's enterprise-level AI hardware DingTalk Real, which can "physically cut off power," the underlying logic is to comply with top-down organizational control.

At this year's Yunqi Conference, Chen Yusen, CEO of Qianwen Office, also expressed a similar idea: "Agents must land in real enterprise scenarios... every task execution must have boundaries and be traceable."

Product design follows the same line. The answer given by Qianwen Office 2.0 is to give Agents the opportunity to intervene in actual execution across departments and positions.

02 AI in the Foreground, DingTalk Behind the Scenes

For a long time in the past, employees handling work needed to first find the corresponding software entrance and switch back and forth among group chats, emails, documents, meetings, and various business systems.

Agents are changing this set of rules. Natural language dialog boxes have become the front end facing users directly, while business systems retreat behind the scenes and become capabilities.

When enterprise services were emerging, employees needed to open ERP and CRM separately to handle work. Later, these scattered entrances were consolidated into one place by IM. Now, Agents are similarly showing signs of swallowing the previous generation of software entrances.

Similar tension also exists in the relationship between Qianwen Office and DingTalk. If a team is already used to communicating and handling affairs in DingTalk groups, why should it take an extra step to create a new collaboration space in Qianwen Office? Qianwen Office provided the case of Intime Department Store as a sample at the launch event.

The launch event introduced that live-stream coupon cards are a business of Intime Department Store with annual transaction volume exceeding 200 million yuan, covering more than 62 stores and involving more than 200 collaborative personnel. However, the daily communication of organizational members is scattered across more than 60 DingTalk groups.

Under the original way of working, a group-buying coupon from submission to listing requires the collection of about 50 pieces of information such as name, price, validity period, applicable stores, and materials. Five operations staff had to browse back and forth among more than 60 groups every day, urge the completion of materials, and then manually summarize them into tables.

Group chats are essentially single-column information streams designed for instant communication. Once business proliferates into dozens of groups, AI Q&A within a single group is unlikely to help much. Cross-group information verification and progress tracking are the real pain points.

In internal testing, Intime linked these more than 60 store groups and the coupon AI table into the collaboration space of Qianwen Office, with the Agent summarizing progress across groups and updating the AI table.

To achieve this kind of cross-group takeover, Qianwen Office at the underlying level uses command-line tools and protocols to break down DingTalk's group chat messages, documents, AI notes, and even Yida applications into atomic capabilities that can be directly scheduled.

In the past, people worked in DingTalk and conveniently invoked AI for assistance; Qianwen Office 2.0 attempts to reverse this so that people watch business results in the Agent space, while DingTalk retreats behind the scenes and becomes a communication pipeline and database silently invoked by agents.

Furthermore, Qianwen Office not only heavily inherits DingTalk's decade of accumulated understanding of organizational processes, but also attempts to take over from DingTalk the task entrance for high-frequency business.

When Qianwen Office can also create groups, collaboratively edit documents, and manage tasks, will DingTalk be sidelined? At present, the official narrative given by Qianwen Office is not one of replacement, but the wording is relatively vague.

Organizationally, both Qianwen Office and DingTalk are headed by Chen Yusen; at the product level, both entrances exist simultaneously.

Shu Junliang, vice president and head of product research and development at Qianwen Office, once said in an interview that traditional office software will not disappear, but will "retreat into a data foundation and context foundation, precipitating those solidified business processes that cannot be generalized." Old software and AI should be in a cooperative relationship.

As for Qianwen Office and DingTalk, he also retains an open judgment on the final product form. Qianwen Office and DingTalk may ultimately become one product, or they may each find their ecological niche. Everything depends on the development of the times and customer needs.

03 Using the Organization as the Unit to Unlock Payment

Whether letting Agents parasitize IM or sideline IM, the direction of focusing on organizational intelligence is very clear. That is, to break the growth ceiling of traditional enterprise software selling seats per capita and unlock enterprises' willingness to pay for organizational intelligence.

The aforementioned office Agents have long been exploring commercialization on the path of individuals and seats, with business models based on credit quotas and subscriptions. But from the perspective of enterprise procurement, tools that merely serve personal productivity can hardly unlock organization-level budgets.

As for basic AI functions such as group chat summaries, meeting minutes, and search, they have also evolved into routine capabilities of collaboration platforms. It is difficult for enterprises to continuously increase large investments merely because employees can write weekly reports faster.

Letting the individual belong to the individual and the organization belong to the organization currently appears to be a relatively reasonable solution.

Not long ago, before the 2.0 version was launched, Qianwen Office cooperated with Taotian to have Li Jiaqi livestream sell a basic membership for 1 cent in the first month. For individual users, efficiency improvement does not seem to be a particularly urgent need, and the entire personal productivity market is still in the stage of building awareness.

To earn big money from enterprise accounts, major tech companies must present another script.

The Doubao Work Partner Enterprise Edition is billed by annual points; the Tencent WorkBuddy Enterprise Edition and the WeCom joint suite introduce team shared credits; Qianwen Office 2.0 also unveiled a pricing plan combining seats and a shared credit pool.

It is not difficult to see that all players are promoting a model of seats plus shared usage. In addition to features, the charging method is also moving closer to enterprise organizations.

In the short term, due to organizational relationships and the accumulation of historical data, the IM fundamentals of Feishu, WeCom, and DingTalk remain solid, and additional purchases and renewals by existing customers are the primary battleground for all parties. However, with the evolution of independent workbenches and cross-platform invocation capabilities, enterprise budgets at the Agent execution layer are opening up new competitive space.

At present, various office Agents are still competing with themselves within their respective ecological territories, trying to achieve additional purchase conversion among existing customers. By next year, we may see competition more intense than the former collaborative office wars.

The copyright of the article cover image and accompanying images belongs to the copyright owners. If the copyright owner believes that their works are not suitable for everyone to browse or should not be used free of charge, please contact us in a timely manner, and this platform will immediately correct it.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.