Tencent's enterprise AI assistant WorkBuddy reached 32 million monthly users by this fall, more than its rivals combined, according to a Bloomberg profile of Tencent Cloud chief executive Dowson Tong published October 29. The figure is the first public scale number for a product that Tencent had treated as a side project inside its internet portfolio, and it lands as the company reframes its AI strategy around business agents rather than consumer chatbots.
The backdrop Bloomberg sets is unusually candid. Founder Pony Ma described Tencent's AI effort this summer as "a leaky ship," a remark that reverberated through a company that had dominated Chinese consumer internet with WeChat and then watched DeepSeek and a wave of AI-native startups set the pace after ChatGPT. Tong's answer was to stop competing for the consumer chatbot and retool around businesses, putting WorkBuddy at the center. The product combines Tencent's existing messaging and office tools with agents that ingest data and generate reports, the same shape as Anthropic's Cowork.
Tong is now one of a handful of executives leading the push to close the gap with ByteDance and Alibaba, with Tencent's standing against DeepSeek, Moonshot AI and other model-native firms at stake. Speaking in Shenzhen this week, he framed the choice as a sequencing decision: enterprise agents first, because that is where the technology can be monetized, with consumers later.
The numbers Tong presented at the Tencent Digital Ecosystem Summit on October 29 go further than the user count, and all of them are company figures. He said the enterprise edition of WorkBuddy has been deployed across more than 50 industries and covers 50 percent of China's Fortune 500 and 60 percent of its top 100 companies; that the open platform has more than 50,000 partners and developers with over 200,000 skills listed; and that international active customers doubled over the past year with overseas AI demand growing at a triple-digit rate across more than 150 countries.
The customer proof point was Yili, the dairy group. According to Tong's presentation, WorkBuddy aggregates business data, generates a sales strategy and pushes it to more than 1,000 frontline store guides, which the company says lifted operational efficiency by over 50 percent. A separate "shopping guide agent" built on Tencent Cloud's Agent Development Platform then generated product recommendations and talking points in real time, which Tencent says raised order conversion by 26 percent in an A/B test. Neither figure has been independently audited, and the definitions of efficiency behind the first one are not public.
The product direction is consolidation. Tencent is wiring its Agent Development Platform into WorkBuddy so that agents built for specific business processes can be published into the assistant and called by employees, alongside a new "Buddy One" suite that bundles Tencent Docs, Tencent Meeting, a newly launched Tencent Drive positioned as memory storage for agents, and search. WorkBuddy is also being pushed into the WeCom sidebar, so staff can assign tasks inside the messaging app and receive results back.
The model layer is next. Tong said Hunyuan Hy4 and a cheaper Hy4 flash are both coming soon, with Hy4 adding visual understanding for the first time and strengthening long-horizon, production-style tasks along with chain-of-thought efficiency. Tencent's own framing of the market is the token volume: China's daily token calls rose from about 100 billion in early 2024 to more than 500 trillion by July 2026, according to figures Tong cited from the National Data Administration.
What the 32 million figure actually proves is narrower than it sounds. Monthly users of an assistant bundled into WeCom and Tencent Docs is not the same as paying seats, and Tencent did not disclose revenue, retention or how many of those users run agent workflows rather than chat. The strategic bet is legible, though: if the entry point to work software shifts from opening an app to instructing an agent, the company that already owns the messaging layer has an advantage that does not depend on having the best model. Tencent is now testing whether that advantage converts into money.
Comments (0)
Log in to join the discussion
Log InNo comments yet