Analyst data from iResearch《China Office Agent Platform Market Research Report 2026》shows that as of May, WorkBuddy’s monthly visits on the PC side reached 8.85 million, leading the second place by 2.6 times, with a quarter-over-quarter growth rate of 72.2%. Tencent’s 2026 Q1 financial report for the same period indicates that, based on daily active account numbers, WorkBuddy has become the most popular efficiency AI agent service in China.
WorkBuddy’s Product Path Evolving from CodeBuddy: a 0.01 Version Built After Two All-Nighters
WorkBuddy’s starting point was not a predefined product roadmap. Its predecessor was Tencent Cloud’s AI coding assistant CodeBuddy; the turning point came when non-technical employees at Tencent Research Institute began to spontaneously use CodeBuddy to search papers and organize content, which has been documented in a deep report by The Beijinger (Ti Media). The team lead, Wang Shengjie, inferred: “Coding is just a process—the end product is the real goal.”
In mid-January 2026, one weekend, Wang Shengjie and an operations colleague stayed up through two all-nighters to complete the WorkBuddy 0.01 version: a minimalist conversational interface plus pre-installed curated Skill packages, with no need for setup wizards or a command-line interface. Before the public beta, more than 2,000 non-technical staff employees within Tencent were already using it daily; multiple media outlets reported and cross-validated this figure on March 9, the public beta launch day, including Guangzhou Daily’s Xinhuancheng. On the public beta launch day, request volume far exceeded CodeBuddy, triggering an emergency tenfold compute scaling.
Three Design Decisions: Natural Language Replaces Technical Concepts, Scenario Templates in Advance, and Ecosystem Native Instead of Plug-in Integration
The differences between WorkBuddy and the Codex route concentrate on three design decisions:
Natural language replaces technical concepts: when users input something like “Help me organize last week’s sales data into a comparison table by region,” the system independently breaks it down into data retrieval, cleaning, analysis, table generation, and output. The interface doesn’t display terms such as “agent dispatch,” “tool calling,” or “context management.” The interaction starting point for Codex and Claude Code is terminal commands or IDE plugins; they need to understand model behavior and manage token budgets—interaction logic designed for engineers.
Scenario templates packaged in advance: Xinjing Bao’s public beta coverage revealed that WorkBuddy has more than 20 built-in Skill packs, covering data processing, ticket processing, document archiving, competitor research, content creation, public sentiment analysis, and sales insights. OpenAI only rolled out its first set of six role plugins in Codex on June 2, 2026.
Ecosystem-native rather than plug-in style integrations: Ti Media reported that WorkBuddy’s integration with Tencent Docs is not done through API calls, but instead “moves in”; it also supports direct remote control via WeChat, so tasks initiated on the computer can be viewed on the phone along with progress updates and additional instructions.
OpenAI Codex: the Turn Path from 20% Data from Non-Developers to Anthropic Claude Cowork
OpenAI’s June 2, 2026 announcement data: Codex’s weekly active users surpassed 5 million, with non-developers accounting for about 20%; non-developers’ growth rate is more than 3 times that of developers. On the same day, six role plugins were launched, covering data analysis, creative production, sales, product design, public equity investments, and investment banking.
The first paragraph of Anthropic Claude Cowork’s official product page admits its origin: the company’s internal non-technical teams first used Claude Code spontaneously, which led to Cowork. Claude Cowork does not provide prebuilt templates; it focuses on local file operations in a desktop application, with no deep native integrations like Tencent Docs.
WorkBuddy Pricing Tiers and Comparison with Key Competitors
WorkBuddy Personal edition: Lite 39 yuan/month, Standard 99 yuan/month, Pro 299 yuan/month; enterprise SaaS flagship tier 198 yuan per user per month (confirmed by Tencent Cloud’s official pricing announcement). Codex starts with ChatGPT Plus, 20 USD/month; Claude Cowork also starts from the Pro plan at 20 USD/month (check the official pricing pages of OpenAI and Claude).
WorkBuddy’s enterprise edition was released on June 5, 2026. Reports by Sina Finance and 36kr include “project” features, an enterprise management back end, and 7×24 expert digital employees.
Frequently Asked Questions
Is the claim that “WorkBuddy’s daily active users rank second in the industry by 3-4 times” supported by third-party data?
According to OmniTools, the currently visible source for this figure is a personal retelling by someone who attended an event on a community platform; the specific statistical methodology, product comparisons, and measurement period have not been made public. Third-party audited data only exists in iResearch’s report—WorkBuddy’s PC monthly visits were 8.85 million, leading the second place by 2.6 times, with a quarter-over-quarter growth rate of 72.2%.
Does WorkBuddy have specific numbers for the share of non-technical users?
At present, public channels only provide qualitative descriptions like “significantly improved,” with no specific percentage that can be directly compared with Codex’s 20% non-developer user share. OmniTools mentioned this qualitative point in a June 16 news brief, but precise numbers have not been released yet.
Is WorkBuddy competing with Codex, or are they targeting different markets?
Based on OmniTools’ analysis, the two have fundamental differences in positioning: WorkBuddy enters the domestic non-technical user market through scenario packaging; Codex enters the global developer market through underlying capabilities, and non-technical user growth is outside the expected growth. There is competition between the two in the non-technical user market for office agents, but their core advantage areas differ.
