Tencent Upgrades WorkBuddy Enterprise Edition Open Strategy, Buddy One Connects Office Suite

Tencent's Agent ecosystem is building an AI that "sees the whole picture." Phoenix Tech Download the app for exclusive early access October 30, 2026, 11:08:22 From Beijing Updated October 30, 2026, 12:13:56

Summary:

Tencent WorkBuddy Enterprise Edition's open strategy is upgraded, and Buddy One connects Tencent's office suite. Competition among office Agents is shifting from "whose model is stronger" to "who can safely see the whole enterprise."

For the past two years, the narrative around AI in the office has been dominated by almost the same sentence structure: help you write emails, take meeting minutes, generate weekly reports. Efficiency is indeed improving, but the reality is that these capabilities are isolated from one another, and AI remains like an intern sitting in the corner of the office—you hand him a piece of paper, he fills it out for you, and then waits for you to hand him the next one.

The problem lies in the fragmentation of "context." How far along a project is, which approval node the budget is stuck at, what the client actually said last time—this information does not automatically appear in the dialog box. It is scattered across chat records, approval flows, documents, and CRM systems, with no connection between them. An AI that cannot see the whole picture can only respond passively; only an AI that can see the whole picture is qualified to talk about "judgment."

From October 29 to 30, the 2026 Tencent Global Digital Ecosystem Summit was held in Shenzhen. The summit released 100+ new products and new capabilities and showcased more than 500 Agent practice cases. The number is not the point. The point is that Tencent wants to answer a question: can Agents truly move from execution tools to an important part of the organization?

Without full-domain context, an Agent is just a faster typewriter

"The biggest problem for enterprises deploying Agents is not that the models are not strong enough, but that there are too many breakpoints," said Dowson Tong, Senior Executive Vice President of Tencent and CEO of the Cloud and Smart Industries Group, bluntly in his speech. "The key to Agent deployment is connection—connecting people, tools, systems, and data, enabling large models to understand business context and truly enter the workflow."

WorkBuddy Enterprise Edition upgraded and released its open strategy at this summit, setting its goal as "helping enterprises land AI productivity." If this sentence is broken down, the real variable is not the words "productivity," but the modifier before "landing": enterprise.

Dowson Tong disclosed that since its launch in March this year, WorkBuddy Enterprise Edition has been deployed in more than 50 industries, covering 50% of China's Fortune 500 companies and 60% of China's top 100 companies; it has more than 50,000 open platform partners, more than 200,000 Skills listed, and has completed a cumulative 9 billion task requests.

If the personal edition Agent solves "single-point efficiency improvement," the enterprise edition Agent has to face a different set of problems: how to manage permissions, how to control costs, where data flows, and who is responsible when something goes wrong.

Tencent's solution is not to build a stronger dialog box, but to put WorkBuddy into WeCom on one hand and use Buddy One to connect Tencent's office suite on the other.

Buddy One connects Tencent Docs, Tencent Meeting, Tencent Drive, and ima, sharing the WorkBuddy Agent foundation, unified accounts, and a unified credit pool, so that users can search authorized content across applications with a single question; Tencent Drive serves as the memory space of the Agent era, uniformly storing the context and generated results accumulated by the Agent and precipitating them as organizational assets.

Through unified accounts, a unified credit pool, and a unified data pool, large models call different tools to generate multimodal content for different needs, and teams can collaboratively edit within projects, jointly review outputs, and continuously discuss and iterate. In this way, humans and AI creating together is about collaboration around tasks, not around file circulation.

Allowing WorkBuddy to "move in" to WeCom as an intelligent robot means that what the Agent reads is not a document manually uploaded by the user, but messages, documents, and meeting records within the authorized scope. Enterprise administrators can configure what AI can see, what it can do, and which steps must be manually confirmed. After AI produces results, they are automatically sent back to the WeCom conversation or approval flow and enter the next stage of the workflow.

Tencent hopes to use this to turn the Agent from an add-on tool into a node within the organization, but whether the node can stand firm depends on whether enterprises are willing to hand over approval flows and group chat records to it. In other words, the premise for this logic to hold is that enterprises believe the benefits of "giving the Agent more context" outweigh the risks.

Context does not only come from inside the enterprise. Dowson Tong mentioned that when a WorkBuddy task needs external information, Yuanbao Search will provide real-time data. Relying on authoritative domestic and international websites, professional data resources, and Tencent's content ecosystem, Yuanbao Search has accumulated a hundred-billion-level index to meet in-depth research needs; it not only serves as a ChatBot providing fast and accurate answers, but also opens search capabilities to the industry through APIs.

Of course, obtaining context is only a ticket to entry. What enterprises really care about is another set of questions: where the money is spent, where errors occur, and where slowness is stuck.

The "Observability Center" launched on October 13 attempts to answer these—real-time observation, session tracking, and trajectory details, three layers of capabilities that drill down from macro traffic to the Prompt and model calls of specific nodes. This is the premise for enterprises to dare to put Agents into real business processes.

In response, Dowson Tong said: "These features let customers measure things themselves and figure out clearly where their money is being spent." As for security and observability, he stressed that these are common capabilities of the platform: "The controllability enterprises need, the data security they need, and the traceability they need are in fact all capabilities provided by Tencent's entire Agent product system."

Building Agents: Tencent Doesn't Bet on a Single Hit

Among the many Agent products, WorkBuddy is an obvious hit. But the more worthwhile question is: why was Tencent able to successively roll out a string of products—WorkBuddy, ima, Marvis, DataBuddy—in less than a year, rather than stopping after one success?

Breaking it down, there are three characteristics that don't quite look like "strategic planning."

First, every starting point was "someone was genuinely stuck." The prototype of WorkBuddy wasn't written in a project proposal—the team discovered that a large number of non-technical employees were using CodeBuddy to handle office tasks but were blocked by the barrier to using an IDE; what they wanted was to get the work done, not to learn programming first. So a product manager built an early prototype over a weekend and gave it to more than 2,000 non-technical employees to use repeatedly, and only then did it grow into a formal product; ima's starting point was the QQ Browser team noticing that many people read a large number of articles every day but couldn't find specific viewpoints again; Marvis's starting point was the Yingyongbao team seeing that users who wanted to convert a file format first had to download a tool and then learn how to use it.

These needs were not "holding a hammer and looking for nails"; the nail pricked the foot first, and only then did the inspiration for a new product emerge.

Second, the foundation is reused repeatedly, rather than being rebuilt for every product. CodeBuddy's core evolved into WorkBuddy; WorkBuddy's Harness was encapsulated into the WMA managed service; Tencent Docs directly adopted WorkBuddy's unified Agent core to achieve human-machine dual writing; DataBuddy inherited the Harness and layered on data Skills; and the open platform then lets partners build applications on the same foundation.

Computing power, runtime environment, memory mechanisms, observability and security—these dirty, heavy chores are accumulated once, and the marginal cost of subsequent products keeps falling. The upgrade to CodeBuddy 2.0 is a typical case: 4 engineers, 4 months, and about 99% of the code generated by AI.

Third, openness is not a posture but layered engineering. The model layer allows enterprises to switch foundations and connect self-built models; the Agent underlying capabilities output the Harness through the open platform, so partners don't have to build a runtime framework from scratch; the enterprise work ecosystem connects inward to WeCom, Docs, Lexiang and WeDrive, and outward is compatible with third-party IM and business systems; the application layer lets enterprises develop their own Agents, Skills and digital employees based on Tencent Cloud ADP. With these four layers of openness stacked together, Tencent's role is no longer just "the one who builds Agents," but an infrastructure provider that makes building Agents repeatable and diffusible.

Retail's operating rules, legal statutes and precedents, and trade quotation processes—this knowledge-intensive experience used to rely on people to provide services, and now it can be packed into Agent applications and Skills, becoming replicable "AI products."

And the direct manifestation of this open strategy at the product level is handing the choice back to customers. "Which model to choose, which Agent to connect, which IM to use—the choice is all in the hands of enterprises," Dowson Tong stressed in his speech.

"Many of our customers simply don't want to be strongly bound to a particular model when building a Harness or an Agent," Dowson Tong further explained in a group interview after the event. "Models iterate very fast; a ranking may be refreshed every two weeks. Each model makes some relative optimization for different tasks, and prices differ. Letting users choose means different models can be used for different tasks. This is a strategic direction we will continue to adhere to."

Models Can Be Swapped, but Organizational Relationships in WeCom Cannot

WorkBuddy supports open compatibility with multiple models; users can use the latest model on Day 0, and enterprises can also connect self-built models. But if the foundation can be replaced at any time, where exactly is WorkBuddy's moat?

In Tencent's account, it is not "the strongest model" or "locking down data," but an open ecosystem that connects capabilities and data Tencent does not have, a platform that provides basic capabilities for partners to build industry applications, multiple models that keep customers from being strongly bound, and context acquisition and authorization architecture that link Agents together.

"In AI, a lot of the time the challenge or limitation lies in whether you can obtain the relevant context, and fully piece together the information needed for a task—your goals, metrics or paths—so that the large model knows how to do it," Dowson Tong believes. From this dimension, he said, "actually everyone's starting line is about the same."

In fact, the long-term moat of office Agents is not the model, but context and ecosystem. No one will give up the entire set of chat tools and organizational relationships the company already uses just to use an upper-layer Agent. The organizational structure, historical chat records and knowledge base accumulated in WeCom are the foundation for office Agents to understand the business. Rebuilding this infrastructure just to switch to another Agent is almost impossible.

This explains why Tencent adopted a "native-level" rather than plug-in connection in linking WorkBuddy with WeCom: identity interoperability, content interoperability, collaboration interoperability, AI capability interoperability, and business management interoperability—five layers of full-chain integration.

Agent access to enterprise context must go through enterprise authorization, and permission and data management still belong to the enterprise. Compared with the industry practice in which "an employee's personal authorization is enough to obtain enterprise IM context," this path is heavier and slower, but it is the prerequisite for enterprise customers to be willing to pay.

Dowson Tong also admitted that the challenge of enterprises deploying Agents often lies on the organizational side: "What I hear more from enterprises today is that they really want to use it, but the key is how to build a secure environment. Because everyone is also afraid that if the Agent just grinds away executing on its own, will it go off track? Will it do something inappropriate?"

The competitive landscape also confirms this. ByteDance is integrating collaborative office tools and migrating from ChatBot to Agent, Alibaba is emphasizing enterprise context, and the three major tech giants each have different emphases. But Tencent's unique card is that it simultaneously owns the office intelligent agent with the most users in China and the collaborative office platform with the largest server user scale, and the two are connected through a native account and permission system, not a shallow API-level connection.

Currently, competition in enterprise-level Agents is shifting from "whose model is stronger" to "who can let AI safely see the full picture of how an enterprise operates." The former is a technology race; the latter is organizational engineering. Technology can be caught up with; the migration cost of organizational relationships is hard to overcome.

The discussion of AI office work has moved from "can it write" to "can it do," but what truly determines the stratification is whether an Agent can see the full picture of how a company operates. An AI that cannot see the whole picture, no matter how fast, is just another input method; only an AI that can see the whole picture has the ability to become part of organizational capability. Tencent is betting on the latter.