How Tencent WorkBuddy Reached 20.97M Visits—and Changed China’s AI Office Race

WorkBuddy’s rapid rise is best understood as a distribution-and-productization story. Tencent launched the desktop workplace agent on March 9, 2026; it recorded 8.85 million PC-side visits that month and 20.97 million in June—roughly 2.4 times as many. In the June ranking of 17 Chinese desktop AI-office agents, Trae recorded 12.79 million visits and QoderWork 7.88 million, putting WorkBuddy ahead of the two combined. 279

That result is significant, but it does not establish that WorkBuddy has already won enterprise AI. The ranking measures PC-side web visit volume and excludes mobile use; it does not measure daily active users, paying companies, revenue, or retention. 2

The growth came from an immediately usable agent layer

WorkBuddy did not ask users to choose a single foundation model or build every workflow from scratch. At launch, Tencent positioned it as a desktop agent for workplace tasks, with support for OpenClaw skills, the Model Context Protocol (MCP), and more than 20 skill packages for work such as research, report creation, and content drafting. It could also switch among Hunyuan, DeepSeek, GLM, Kimi, and MiniMax models. 42

This matters because an office agent needs more than a capable model. It needs a practical way to turn requests into repeatable work. OpenClaw compatibility gave WorkBuddy a familiar skills format and a path to reuse an existing ecosystem, while packaged skills reduced the setup required before a user could automate a task. That is a plausible adoption advantage, though public reporting does not isolate how much traffic each feature generated. 2442

From chatbot to workplace orchestration

The product proposition was broader than asking a model questions. Tencent added a skills center, expert mode, scheduled automation, and a knowledge-base capability, while integrating Tencent Documents and supporting task delivery through WeChat and WeCom. 28

Reports also describe complex tasks being split into subtasks handled by multiple agent instances. In practice, that approach shifts the product’s value from a single model response toward the system that coordinates research, drafting, data work, files, messaging, and follow-up actions. 46

Multi-model access reinforces that shift. Users can choose or switch models according to a task and cost rather than commit their workflow to one proprietary model. Tencent’s published pricing offers a free tier alongside RMB 99, RMB 199, and RMB 999 monthly plans, with paid tiers offering full model selection. 43

Tencent’s ecosystem made distribution a strategic advantage

WorkBuddy was built on Tencent Cloud’s CodeBuddy foundation, rather than created as an entirely new agent stack. Public reporting also describes integrations with WeChat, WeCom, QQ, Tencent Documents, Tencent Meetings, and other workplace tools, including the ability to issue or execute tasks through messaging channels. 192847

Those links are strategically important because office software gains value when it can reach the places where work already happens: communication, documents, identity, devices, cloud services, and enterprise administration. Tencent’s PC-oriented channels and cloud footprint could reduce the friction of discovery, installation, and deployment compared with a standalone agent product.

That does not prove the integrations alone caused WorkBuddy’s traffic lead. It does show why the contest cannot be judged only by model benchmarks: the product that owns more of the workflow surface can make a comparable model more useful in daily work.

Hunyuan upgrades and pricing reduced friction—but were not the whole story

Tencent’s Hunyuan updates strengthened the company’s in-house model option. Hy3 became available through WorkBuddy, and Tencent offered it free to global WorkBuddy users through August 31, 2026. 51

Still, WorkBuddy’s strategy was not model exclusivity. Its ability to offer Hunyuan alongside DeepSeek, GLM, and Kimi lets the workplace layer remain useful when users prefer another model for a specific task. The credit-based plans and free entry point similarly lower the cost of experimentation before a user or team commits to larger usage. 4345

The likely implication is that Hunyuan helped improve the underlying capability and economics of the platform, while model choice helped WorkBuddy compete for the user’s workflow rather than demand loyalty to one model.

Why competitors began consolidating

The rapid traffic concentration coincided with organizational changes across the market. Alibaba consolidated QoderWork, Wukong, and MuleRun into Qianwen Office, which entered public beta in August. Reports also described ByteDance bringing its AI-office efforts into closer alignment across Feishu-related and other product lines. 12

These moves are consistent with a market moving away from fragmented point tools. Coding assistants, office copilots, cloud-operation tools, and chatbots can each solve pieces of work, but an agent workspace is more compelling when tasks, files, communication, models, and governance are connected.

What WorkBuddy still has to prove

The strongest public number—20.97 million monthly PC-side visits—is an adoption signal, not a complete business metric. Visit totals can include repeat use and do not reveal how many organizations deploy the product broadly, how many users pay, whether teams expand usage, or whether they remain after trial incentives end. 27

Durable workplace network effects would require evidence that teams build and share skills, templates, permissions, organizational context, and accumulated work history inside WorkBuddy. Those assets could raise switching costs over time, but they remain a hypothesis until Tencent reports stronger retention, paid-seat, collaboration, and revenue data.

For now, WorkBuddy’s rise shows that China’s AI office-agent race is increasingly about the layer above the model: workflow design, skills, integrations, deployment, and distribution. Tencent moved quickly to assemble that layer around an accessible desktop product—and the June traffic data shows the approach found a large early audience. 242