Tencent Releases WorkBuddy Enterprise Edition and Buddy One, Proposing Four Hurdles for Enterprise Agent Deployment

Statistics from the National Data Administration show that China's average daily Token calls grew from about 100 billion in early 2024 to more than 500 trillion in July this year, an increase of more than 5,000 times in two years. A Token is the smallest unit of information processed by AI, and behind every call are the computing power, electricity and costs that enterprises actually pay.

The prosperity at the individual level needs no argument, but the picture at the enterprise level has always been blurry. A growth of thousands of times means it is no longer a trial taste of something new, but a sizable and still expanding investment. But how to evaluate and measure it still has no generally accepted answer.

On October 29, the 2026 Tencent Global Digital Ecosystem Summit opened in Shenzhen. In his keynote speech, Dowson Tong, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industries Group, offered this explanation: the biggest problem for enterprises deploying Agents is not that the models are not strong enough, but that there are too many breakpoints—if AI is to do analysis and write proposals but cannot connect to the enterprise's internal systems to get data, it has no way to understand the actual situation of the business, channels and customers.

Tencent Group Vice President and Tencent Cloud CTO Wang Huixing went further, pointing out that context connection is only the starting point. For enterprises to truly realize the value of Agents, they also need to solve problems such as business integration, security governance, effect measurement and large-scale deployment, which is exactly what WorkBuddy is doing.

As the operator of China's largest AI office agent by user scale, Tencent's deployment answer can be glimpsed through WorkBuddy. Over the past eight months, this product has moved from a personal assistant to an enterprise system, and almost all the new progress announced at this summit is in this direction.

The threshold has shifted from models to context, and even to the level of organizational management

Efficiency gains at the individual level have been confirmed by data. Analysys data shows that in September 2026, it ranked first among China's desktop AI-native office agent platforms with 85.34 million monthly PC visits.

But at the enterprise level, this change has only just begun. On the matter of "getting AI productivity to work," Dowson Tong offered a testable formulation in a conversation on the afternoon of the 29th. He first asked in return: can you still work without AI today? "I think it's quite hard to go back, so whether it works—the first thing is the value it creates, whether you are willing to give it up. If you can't give it up, then it has already become part of our work and part of our lives."

In his view, if an enterprise has not grasped the fundamentals and has not adjusted its organization to enable employees to use AI well, "then you may pay the same cost as others, but when you don't get the same return as others, naturally it won't feel like it's working for you."

The same Tokens and the same tools—the output depends on whether the enterprise is willing to hand over its processes, permissions and judgment criteria as well. Therefore, the observability center launched in the WorkBuddy Enterprise Edition addresses the question of how to measure results. It lists which department and which scenario the Tokens are spent on, and which step is slow or expensive. What enterprises care about has shifted from whether it can be used to whether it can be managed and whether the investment is worth it.

As for exactly where the organizational side is stuck, in conversations around the summit, the answers given by three Tencent business leaders were all about the conditions that must be met before agents enter enterprises. Wang Huixing said that models and Harness have already crossed that threshold in terms of product and technology, and the next task is to prepare context so that agents can access enterprises' real work scenarios and workloads. He ranked opening up platform capabilities first among four major capability upgrades, with the aim of letting Agents connect to enterprise context.

He used an analogy for the difference between individuals and enterprises: individual efficiency gains are like giving each person a clone or assistant to handle point-to-point matters such as emails, schedules and to-dos; back at the enterprise, what is needed is not just one assistant per person, but turning professional roles such as programmers, legal staff and finance staff into callable digital employees. They need to access customer data and enter business systems; if permissions are not controlled and results are not verified, an entire business line is affected.

The scale figures announced at the summit are that the WorkBuddy Enterprise Edition has covered half of China's top 500 enterprises, and that since its launch the product has completed more than 9 billion task requests in total.

To truly connect enterprise systems and data, the boundaries of permissions must first be clearly drawn. Dong Zhiqiang, Vice President of Tencent Cloud and head of security, said that Agent behavior going off track is a real phenomenon; unlike traditional software, model behavior is an open question. The countermeasure is to confine behavior in a sandbox, keep a full audit trail, treat all external materials as data, and route calls to internal tools through a gateway whitelist.

Dowson Tong's speech presented exactly the complete version of this context mainline. Buddy One, released at the summit, is a cross-application collaboration system for individual users. With WorkBuddy as the workbench, it connects products such as Tencent Docs, Tencent Meeting, ima and Tencent Drive, so users no longer need to move files or copy context between applications; a single question can retrieve authorized documents, meeting and knowledge base content.

The official summary of this change is that collaboration revolves around task coordination rather than file circulation. File circulation only requires format interoperability, while task coordination requires context interoperability.

Context is distributed across several different places. What documents, meetings and knowledge bases accumulate is what the enterprise has done; what daily communication conveys is what it is doing; and what ERP, CRM, OA and financial systems store is how the enterprise operates. The first two categories rely on connections between products, while the last can only rely on open system interfaces. WorkBuddy's approach is to have management systems open MCP interfaces, then use skills to clearly describe the data structures and relationships, so that the Agent can retrieve data and break down tasks on its own.

The simultaneously released Tencent Drive handles unified storage and is positioned as the memory space of the Agent era, with enterprise edition storage expanding from 5GB to 50GB per seat. Combined with deep integration with WeCom, employees' communications, documents and meeting records in WeCom become, within the scope of authorization, the basis for the Agent to understand tasks, and enterprises do not need to change systems, move data or redo processes.

Whether this path can be accepted by more enterprises also depends on whether it locks them in. Tong Daosheng said in the discussion that Skills themselves can be taken away; they are merely text and descriptions of data interfaces. As a product provider, of course one hopes the product has stickiness, but to truly serve a product to a large number of users, an open ecosystem is one way, one that can connect many capabilities and data that Tencent does not possess.

Wang Huixing added that while WorkBuddy is deeply co-built with Hunyuan, it also connects to mainstream open-source models such as KIMI and supports enterprises' self-deployed models. The right to accumulate data, context and memory remains on the enterprise side, and only then are enterprises willing to hand over their real business context.

The deeper it is used, the thicker the enterprise context accumulated in the platform, and the more closely the agent's delivery fits the enterprise's own standards; the premise is that the enterprise is willing to hand over more of its daily work to it.

There are no shortcuts in organizational change

In his speech, Wang Huixing summarized the thresholds an enterprise must cross to implement Agents as four gates: it must run, be trustworthy, be measurable and be easy to implement. Only when it can be connected can it be managed; only when it can be managed does one dare to calculate investment; only when it can be calculated clearly is one willing to replicate. The first three correspond respectively to open platform capabilities, enterprise-grade security solutions and an observability center, and the last gate is replication: whether an agent validated by one team can be pushed to more teams and scenarios. After the observability center went live, Suning.com used it to adjust prompts and model configurations, reducing Token consumption by 10%.

The hardest is the first step, making it run. A leading manufacturing enterprise proposed 13 scenarios in after-sales, product and marketing, and WorkBuddy directly connects to its dealer management system through MCP, automatically pre-reviewing 20,000 claim invoices per month. The real difficulty lies on the human side. Previously, more than a thousand employees had spontaneously used the personal edition, and the management team was caught between banning it and letting it be. On the second day after the enterprise edition's unified controls went live, more than 2,000 test accounts were claimed by employees, and the number of AI users within the enterprise expanded more than tenfold. Connecting to systems is a technical issue; bringing thousands of spontaneous users under unified control is an organizational issue.

Among WorkBuddy's customer cases, there are three comparable samples, from daily chemicals, telecommunications and dairy respectively.

Zhongshun Jierou completed mandatory WorkBuddy assessment training for 1,200 people in October this year. Jierou CDO Yang Senlin gave the supply chain as an example. He disclosed that the entire chain from sales orders, shipping and warehouse selection to production scheduling is handed to the agent to seek the optimal solution, and cost savings reached the four-digit level. After business staff took matters into their own hands, the rules originally scattered among outsourced teams and the IT department began to accumulate within the enterprise.

Unicom Huasheng is a wholly owned subsidiary of China Unicom. In the invoicing scenario, finance staff at 31 provincial companies each used to spend two days a month reconciling three tables: payments, orders and invoices. Now the three tables and matching rules are handed to the agent, and one finance staff member can complete it in two hours a month. Vice General Manager Chen Fengwei introduced that, based on a summary of about 100 internal cases, the efficiency improvement for repetitive work is about tenfold.

The third example Tong Daosheng cited in his speech was Yili, in front-end businesses such as marketing and sales. Yili combines Tencent Cloud ADP, WorkBuddy and WeCom: WorkBuddy aggregates business data and generates business strategies, which are then distributed to thousands of shopping guides, and ADP's shopping guide agent generates recommendations and scripts. According to the data in the speech, operational efficiency improved by more than 50%, and in A/B tests sales orders increased by 26%.

The three samples are in different industries, but what they share is that none of them started over; all gradually brought Agents into existing systems and ways of working.

The difficulty on the organizational side was illustrated by Tong Daosheng with the example of a programmer. Within Tencent, the R&D team was the first to use it. After coding efficiency improved several times over, the pressure of integration and quality assurance fell on the testing team, and deeply hidden bugs and security vulnerabilities might only appear much later.

"The organization must change. What everyone is doing is also changing," he said. This process will sort out who uses it well and who does not, and enterprises will then set requirements accordingly. As for how to change, he did not give a universal solution, saying only that different positions will go through a similar process: first fully embracing the efficiency gains brought by the tools, then discovering new problems, and then the enterprise setting requirements.

After leading in scale, leading in productivity is the real turning point

In terms of scale, WorkBuddy is already ahead in the office agent track. A report released by Analysys in June shows that its monthly visits have exceeded the combined total of the second and third place. Visits reflect breadth of adoption and have no direct correspondence with output.

Leading in scale does not mean competition has weakened. DingTalk, backed by Alibaba, positions AI as the entry point to organizational processes; Feishu, backed by ByteDance, focuses on knowledge collaboration; WeCom's advantage lies in its connection to WeChat. The three companies' entry points are processes, collaboration, and connection, respectively, and ultimately all must prove whether Agents can enter enterprises' core processes.

Facing horizontal competition, Liu Yi's answer is, "We never thought about the word 'defend'."

In his view, the market is still early, more than 90% of enterprises have not yet used Agents at scale, and the incremental market is far larger than the existing one, so current share proves nothing. Tencent also does not lack a foundation: WeCom and Tencent Cloud each hold millions of customers, and WorkBuddy can enter these customers' scenarios and connect their context.

The remaining difference lies in execution, and he compares three things: service, model collaboration, and on-site implementation. He also judges that there will not be a single dominant player in the B2B market, and that ultimately platforms, products, and models will be combined to serve different customer groups separately. The current lead is therefore limited to the adoption level.

How to measure output deep within processes currently has no unified answer. Dowson Tong observed that in past years the industry focused on model parameters and benchmark scores, but this year such discussion has noticeably decreased, and the metric has shifted to "what exactly did using AI in a certain scenario achieve." He also said it is difficult to measure the entire industry with a single number, because costs vary greatly across scenarios; Token consumption can reflect how much cost users are willing to pay for a task, but it is only a rough measure.

At the enterprise level, cost-effectiveness manifests as whether processes can operate as a whole. What will ultimately determine this competition—models, Harness, or the ecosystem—and what conditions will make the inflection point materialize? Wang Huixing's answer is "openness," which is Tencent's DNA and also its advantage.

Wang Huixing believes that enterprises should not tear down and rebuild existing systems just to use Agents, nor should they be locked into a single tool or model. WorkBuddy integrates internally with WeCom and Tencent's office products, and externally connects to third-party collaboration tools and enterprises' own business systems through MCP, connectors, and other means, while providing enterprises with diverse model choices. Only by letting Agents enter real business environments and call the data and tools needed to complete tasks can they truly move from demos to production. (This article was first published on TMTPost APP)