On June 2, 2026, OpenAI disclosed a set of data in an official announcement: weekly active users of Codex had surpassed 5 million, with non-developers accounting for about 20%, and their growth rate was over 3 times that of developers. OpenAI's conclusion was straightforward: "Non-developers are the real story."
At the same time, in the Chinese market, the daily active users of WorkBuddy had already reached 3 to 4 times that of the second-ranked product. This gap began widening in March. The influx of users isn't just engineers; HR, operations, and administrative staff are becoming the main user groups.
Both products are telling the same story: enabling non-coders to use AI Agents. But the paths they take are completely different. Codex and Claude Code start from the command line and IDE, moving towards office scenarios. WorkBuddy starts from office scenarios, packaging Agent capabilities into tools that can be used without a manual. This difference in approach explains why non-technical users flocked to WorkBuddy first.
A Product "Forced" Out by Users
WorkBuddy wasn't a pre-planned item on the product roadmap.
Its predecessor was CodeBuddy, an AI coding assistant developed by Tencent Cloud. Following the normal script, this team should have continued deepening its work on developer tools. But a turning point was documented by TMTPost in an in-depth report: non-technical employees at Tencent Research Institute began spontaneously using CodeBuddy to search for papers and organize content. These people's main jobs were research and report writing, unrelated to coding. Team leader Wang Shengjie saw the meaning in this. His assessment, cited by TMTPost, was: "Coding is just the process; the output is the goal."
During a weekend in mid-January 2026, Wang Shengjie and an operations colleague worked through two overnighters to create the 0.01 version of WorkBuddy. An extremely simple conversational interface, pre-installed with curated Skills, ready to use upon opening. No configuration wizard, no command line, no technical jargon. Before public beta, over 2,000 non-technical Tencent employees were using WorkBuddy daily, a number cross-verified by multiple media outlets including Guangzhou Daily's Xinhuacheng in reports on the March 9th public beta day. The job distribution among these users was broad: HR, admin, operations, sales—none earned their living by coding.
The starting point for this product line wasn't "we think office Agents are a good market," but rather "a group of people who don't write code are already using programmers' tools to make do with their work, let's give them something they don't have to make do with." From day one, WorkBuddy targeted non-technical users, not because market analysis said so, but because user behavior already did.
Three Design Decisions
Wang Shengjie said something in a TMTPost interview: "Users shouldn't need to understand what an Agent is, what a tool is, or what prompt engineering is. They just need to know what they want."
This statement summarizes WorkBuddy's first design decision: replace technical concepts with natural language. In WorkBuddy, a user inputs "Help me organize last week's sales data into a comparison table by region," and the system automatically breaks it down into data retrieval, cleaning, analysis, chart creation, and output. Users won't see terms like "Agent scheduling," "tool invocation," or "context management" on the interface. In contrast, Codex and Claude Code start interaction from terminal commands or IDE plugins, requiring an understanding of model behavior, managing token budgets, and handling execution errors—this is an interaction logic designed for engineers, not for HR.
The second decision was pre-packaged scenario templates. The Beijing News disclosed in its public beta report that WorkBuddy comes pre-loaded with over 20 Skills packages, covering data processing, invoice processing, document archiving, competitor research, content creation, public sentiment analysis, and sales insights. Behind each Skill is a preset workflow; users can open and use it without needing to design an automation process from scratch.
For comparison, it wasn't until June 2, 2026, that OpenAI launched Character Plugins in Codex, with the first batch of six covering data analysis, creative production, sales, product design, public equity investment, and investment banking. Anthropic's Claude Cowork takes a different path; it doesn't provide pre-made templates but lets users directly operate local files and applications on their desktops through natural language. This idea came from an awkward discovery: Anthropic's official product page admitted that internal marketing and data teams were bypassing the chat interface to directly use Claude Code, which is aimed at developers.
The third decision was ecosystem nativization, not add-on integration. TMTPost's report pointed out that WorkBuddy's integration with Tencent Docs isn't through API calls, but by "moving in." Users can directly summon the Agent within Tencent Docs to process the current file, without switching back and forth between two applications. It also supports direct WeChat remote control; tasks started on a computer can have their progress checked, instructions supplemented on a phone, and then resumed on the computer. For domestic enterprises already heavily using WeChat, WeCom, and Tencent Docs, WorkBuddy isn't "another AI tool"; it's an extra function within their existing workflow.
Codex takes a different route. OpenAI's June 2nd announcement listed 62 integrated applications, from GitHub and GitLab to Salesforce, HubSpot, and Snowflake. This is a general integration strategy for the global market, broad enough, but depth depends on the quality of each third-party API. Claude Cowork focuses on local file system and desktop application operations, with less deep native integration into office suites.
These three decisions correspond to the removal of three types of barriers: cognitive barrier (no need to understand technical concepts), scenario barrier (no need to deconstruct task processes oneself), and environment barrier (no need to leave the office software being used). Codex's Character Plugins and Claude Cowork are both moving in the same direction, but WorkBuddy delivered a complete non-technical user solution ahead of schedule in the first half of 2026.
What the Growth Data Says
The change in user structure is ultimately reflected in the numbers.
Analysys gave a clear ranking in its "China Office Intelligence Platform Market Research Report 2026": as of May 2026, WorkBuddy's PC monthly visit count was 8.85 million, leading the second place by 2.6 times, with a month-on-month growth rate of 72.2%. Tencent's 2026 Q1 financial report confirmed this from another angle; DoNews, in its earnings analysis, cited Tencent's statement that in terms of daily active accounts, WorkBuddy had become China's most popular efficiency AI Agent service. CVCNews reported that WorkBuddy's month-on-month growth rate in March reached 831%. During the same period, TMTPost reported that request volume on the public beta day far exceeded that of CodeBuddy, triggering urgent capacity scaling alarms, ultimately leading to a 10x scale-up.
The increase in non-technical user proportion currently only has qualitative descriptions; "significant increase" appeared in an OmniTools news flash on June 16th, with no specific percentage seen in public channels. If available, it could be directly compared with Codex's 20% non-developer share. But currently, the precision of WorkBuddy's user profiling remains weaker than Codex's side.
Nevertheless, the direction is clear. The growth curve highly coincides with the timing of the non-technical user influx. The post-public-beta growth rate in March and the traffic entry effect of Tencent's ecosystem together pushed WorkBuddy to the number one spot.
Overseas Products Are Turning Around
OpenAI and Anthropic both saw the same signal.
The data disclosed by OpenAI in its June 2nd announcement was itself the signal: non-developer growth rate was over 3 times that of developers. This isn't a fringe user group growing; it's a new main growth driver forming. The six Character Plugins and Codex Sites feature released the same day were a clear "non-technical pivot."
Anthropic's story is more straightforward. The very first paragraph of Claude Cowork's official product page admits the product's origin: internal non-technical teams "voted with their feet" by using Claude Code first, leading to the creation of Cowork. Anthropic's conclusion: "Most AI assistants require users to break work into single prompts, Claude Cowork accepts a result and handles the rest itself." Forte Labs pointed out in a comparative analysis that Claude Code requires command-line installation and is aimed at developers. Claude Cowork offers a simplified interface, but its positioning is still independent desktop operation, handling local files and apps, not designed for scenarios like "directly editing reports inside Tencent Docs."
Both products are turning around. But tools that start from CLI or IDEs require not just feature additions to transform the interaction paradigm from command-line to GUI and expand the user model from developers to HR and admin; they also need a reconstruction of permission systems, governance frameworks, scenario templates, and ecosystem integration methods. This switch itself takes time. The window period WorkBuddy gained in the first half of 2026 is precisely the gap during this transition cycle.
Two Paths, Not Which is Better
WorkBuddy's chosen path is to build the Agent inside the user's office software. WeChat, Tencent Docs, WeCom, QQ—these are tools users are already using. The Agent isn't the destination; it's a capability layer embedded into the existing workflow. Codex and Claude Code's chosen path is to build an Agent, then have users come to it. Codex is in the terminal, in the IDE, in the ChatGPT app. Claude Code is even only in the terminal; Claude Cowork is on the desktop. These products have stronger core capabilities, more flexible models; developers can use them to build workflows in any scenario, but non-technical users must first cross the hurdles of installation, configuration, and understanding the interaction logic to "reach" this Agent.
One of Wang Shengjie's assessments: "Users shouldn't need to understand what an Agent is; they just need to know what they want." The difference between this product philosophy and that of Codex and Claude Code is the fundamental divergence between the "scenario encapsulation" and "underlying capability" approaches. The former pursues "shortest path for users to find the Agent," the latter pursues "most things the Agent can do for users."
In the domestic market, WeChat and WeCom cover a much larger user base than the terminal user base; WorkBuddy's starting point itself is closer to non-technical users. In overseas markets, GitHub and the terminal are developers' territory; it's completely reasonable for Codex and Claude Code to start from there. The influx of non-technical users was unexpected growth, requiring time to accommodate.
Pricing also reflects the same difference. WorkBuddy's personal Lite version is priced at 39 RMB/month, Standard at 99 RMB/month, Pro at 299 RMB/month, and the enterprise SaaS flagship version at 198 RMB/person/month, as confirmed by Tencent Cloud's official pricing announcement and third-party channels. Codex requires a ChatGPT Plus subscription starting at $20/month, per OpenAI's official pricing page. Claude Cowork also starts from a Pro plan at $20/month, per Claude's official pricing page. WorkBuddy's entry price is lower, naturally giving it wider reach in the domestic market.
How Long is the Leading Window
WorkBuddy leads by about half a year in "scenario encapsulation," but competitors' steps aren't slow.
OpenAI's June 2nd announcement planned a roadmap for over 11 Character Plugins, with non-technical user growth over 3 times that of developers. Anthropic positions Claude Cowork as "for non-technical tasks," a strategic-level product pivot. Neither team treats office Agents as a side business.
WorkBuddy's enterprise version was released on June 5, 2026, including "Project" functionality, enterprise management backend, and 7×24 expert digital employees. The enterprise version is a key step in securing its advantage, expanding from individual users to organizational procurement, with completely different pricing models and retention logic. The accumulated capabilities in underlying models by Codex and Claude Code are objective facts; OpenAI and Anthropic possess the world's most advanced model training capabilities. When they complete the transformation of the interaction layer, WorkBuddy's first-mover advantage will face more direct competition.
A 46-year-old government cloud salesperson generated a city economic analysis report on WorkBuddy in 15 minutes; TMTPost reported this case. An independent developer set up 6 digital positions, with a monthly cost of 800 RMB replacing 50,000 to 80,000 RMB in labor costs. When non-technical users can have an Agent independently complete work without learning anything, the gap in daily active users isn't an anomaly, but the natural result of the barrier difference.
