Jade GAO

Tang Daosheng, Tencent's senior executive vice-president and chief executive of its Cloud and Smart Industries Group (CSIG), unveiled Buddy One, an integrated suite built around the WorkBuddy AI office agent, at the Tencent Global Digital Ecosystem Summit in Shenzhen on October 29. The suite links the accounts and points pools of Tencent Docs, Tencent Meeting, ima and Tencent Cloud Drive, which was also launched at the event.

Tencent also disclosed that WorkBuddy had processed 9 billion task requests since its launch in March, going through 100 major iterations in seven months. Its enterprise version has been deployed across more than 50 industries, covering 50 per cent of China's top 500 companies and 60 per cent of its top 100. More than 50,000 partners and developers have joined its open platform, which hosts more than 200,000 Skills — reusable skill plug-ins. Tencent also announced that its new Hunyuan models, Hy4 and Hy4 flash, would be launched soon.

Buddy One is being offered with a VIP membership package priced at 700 yuan (about HK$760) per month on a recurring basis. It includes the flagship version of WorkBuddy, the professional version of Tencent Meeting and the Plus super membership for Tencent Docs, as well as AI points usable across products and 1TB of storage. The monthly price of WorkBuddy's individual flagship version was already 700 yuan; the main change under the new package is that membership benefits for several products are combined in one subscription.

It was WorkBuddy's predecessor that nearly got cut

WorkBuddy grew out of a department that once came close to being shut down. In an article published in August, Tang recalled that CSIG closed a number of loss-making SaaS and PaaS products during its cost-cutting and efficiency drive. It also considered reducing investment in its developer tools (DevTools) team, but ultimately kept it. The team later developed the CodeBuddy coding assistant, focusing on AI programming.

WorkBuddy itself was only created early this year. According to Caijing, product manager Wang Shengjie used AI programming to write the initial version in mid-January. It began internal testing at Tencent on January 19, with more than 2,000 non-technical employees taking part. After the open-source AI agent tool OpenClaw — popularly known as "lobster" on the mainland — became a major hit there early this year, Tencent brought WorkBuddy to market ahead of schedule. It officially launched on March 9. Traffic far exceeded expectations after the launch, causing temporary instability and forcing the team to increase capacity tenfold in an emergency.

Tencent significantly stepped up its investment after the launch. James Mitchell, the company's chief strategy officer, said at a second-quarter results conference call that Tencent had sharply increased investment in WorkBuddy after seeing its explosive growth, while cutting promotional budgets for other AI products. Asked whether the investment would continue, Tang said: "Our capital is limited, and no investment is unlimited. But at this stage, it is indeed a highly important strategic product." He added that this was a fast-growing market and that WorkBuddy was currently in a relatively leading position: "This sector is only just getting started."

Monetisation: no numerical targets yet

Tang said WorkBuddy's monetisation was primarily intended to support the development of its products and services, rather than to set a numerical target for the team to meet. The enterprise version currently charges modest fees for security controls, observability and management functions. He described the revenue as "a little", but did not disclose an amount. He said its main purpose was to ensure that third-party partners in the ecosystem were not "working for free".

He said the key measures for assessing the business were whether users stayed after subsidies ended and whether they were willing to pay. However, Interface News reported in August that under WorkBuddy's low-price strategy, user growth could not yet be equated fully with stable demand for paid services. The same test will ultimately be applied to WorkBuddy itself.

Why Tencent is not following Alibaba and ByteDance in merging operations

China's three major technology companies have all restructured their AI office businesses this year. Alibaba established the Alibaba Token Hub business group in March and placed DingTalk under it. In July, it combined three agent products — QoderWork, Wukong and MuleRun — into "Qianwen Office", under the leadership of DingTalk's new chief executive Chen Yusen. On July 30, ByteDance folded the Feishu product team into Doubao, while Feishu's marketing, sales and customer service teams were merged with the cloud business Volcano Engine. The common direction of the two companies is to bundle collaboration tools with models and cloud infrastructure into integrated solutions, simplifying customers' purchasing decisions and strengthening platform lock-in.

Tencent also merged the business of another office agent, QClaw, into the WorkBuddy division in July, but did not combine its office products with its cloud business. Asked whether it would follow Alibaba and ByteDance with further integration, Tang replied: "Not for the time being."

Tencent has chosen an open approach. WorkBuddy can connect to Tencent's own Hunyuan models as well as third-party models including DeepSeek, Zhipu GLM, Kimi and MiniMax. It also supports privately deployed models operated by enterprises themselves. Tang said many customers did not want to be tied to a particular model when using an AI agent. As the various models are improving rapidly and their rankings are frequently changing, he said Tencent would continue to pursue a multi-model strategy.

Asked whether enterprises would be able to take away the Skills they had built if they later left WorkBuddy, Tang said plainly that they could, because Skills were essentially just text descriptions. "For a product to genuinely serve a broad user base and meet all kinds of needs, an open ecosystem is one way to do it. It allows us to bring in capabilities Tencent does not have and enables our partners to build other services," he said.

The two approaches involve different trade-offs. Alibaba and ByteDance's strategies are closer to the traditional SaaS model: they use an integrated experience to raise customers' switching costs in exchange for higher retention and average spending per customer. Tencent is closer to the logic of the early internet's open platforms, using low barriers to attract participants before monetising through scale. But openness is a double-edged sword: Skills can be taken away, meaning enterprises can migrate at any time if a rival offers a better experience. Tencent will need to strike a balance between being "open enough not to be accused of lock-in" and retaining sufficient stickiness. The outcome will ultimately depend on user retention and ARPU, or average revenue per user.

9 billion tasks do not equal a 9 billion yuan business

Tang said there were two levels to assessing whether AI productivity had "worked": whether users were willing to give up the value created by AI, and whether the cost savings, efficiency gains and additional revenue generated after enterprises paid for Tokens could cover the cost. Tokens are the units used to measure an AI model's computational consumption.

There have been doubts overseas about the return on investment in AI. Tang said mainland models had consistently focused on value for money, while the various open-source models were making intelligence cheaper and lowering the threshold for enterprises to try the technology. He cited companies that had used AI to improve marketing efficiency, generating more gross merchandise value (GMV) with the same marketing spending.

Tang also said that although Token consumption was not a rigorous measure, it could roughly reflect how much users were willing to spend on a task. It should be noted that the 9 billion figure disclosed by Tencent refers to the number of "task requests", not Token usage and certainly not revenue. It also includes a large number of individual users paying with points. The figure reflects usage activity and cannot be directly converted into enterprises' willingness to pay.

What to watch next

First, revenue contribution. Since its second-quarter results, Tencent has disclosed operating profit excluding the impact of newer AI products including Hunyuan, Yuanbao, CodeBuddy, WorkBuddy and Xiaowei. Second-quarter non-IFRS operating profit rose 19 per cent year on year to 86.1 billion yuan (about HK$94 billion). But WorkBuddy's standalone revenue has not been disclosed. Whether the next quarterly results can provide more detailed figures will be key to assessing its monetisation progress.

Second, ecosystem quality. The 200,000 Skills are merely a measure of quantity; more important is how many high-quality agents are used repeatedly.

Third, competitors' moves. This includes whether Qianwen Office and Doubao will follow with support for multiple models, how they will adjust Token pricing, and competition among the three companies in overseas markets.

Fourth, organisational management. Tang acknowledged that the parallel development model used by WorkBuddy's small teams in its early days enabled rapid iteration. But the teams ultimately have to merge their versions, and without an overarching architectural design, different functional modules could "step on each other". He added that the industry was still exploring new organisational models brought by AI and had no clear answer yet.

When every company can offer a reasonably good AI office experience, the real barrier may not be technology, but which platform ultimately becomes embedded in an enterprise's working habits.