On June 2, 2026, OpenAI disclosed some data in an official announcement: Codex weekly active users exceeded 5 million, with non-developers accounting for about 20%, and the growth rate was more than three times that of developers. OpenAI's judgment was straightforward: "Non-developers are the real story."

At the same time, in the Chinese market, WorkBuddy's daily active users were three to four times that of the second-ranked product in the industry. This gap began to widen in March. It wasn't just engineers joining; HR, operations, and administrative staff were becoming the main user groups.

Both products tell the same story: enabling non-coders to use AI agents. However, they take completely different paths. Codex and Claude Code start from the command line and IDE, targeting office scenarios. WorkBuddy, on the other hand, starts from the office scenario, encapsulating agent capabilities into a tool that can be used without instructions. This difference in approach explains why non-technical users flocked to WorkBuddy first.

A product "forced" out by users

WorkBuddy is not a default item on the product roadmap.

Its predecessor was CodeBuddy, an AI code assistant developed by Tencent Cloud. Normally, this team should have continued to focus on developer tools. However, a turning point was documented in an in-depth report by TMTpost: non-technical employees at Tencent Research Institute began spontaneously using CodeBuddy to search for papers and organize content. These people's primary jobs were research and report writing, unrelated to coding. Team leader Wang Shengjie saw the significance of this. His assessment was quoted by TMTpost as: "Coding is just a process; the product is the goal."

On a weekend in mid-January 2026, Wang Shengjie and an operations colleague worked two all-nighters to create WorkBuddy version 0.01. It featured a minimalist interface, pre-installed selected skills, and was ready to use immediately. There were no configuration wizards, no command lines, and no technical jargon. Before the public beta, over 2,000 non-technical employees within Tencent were already using WorkBuddy daily, a figure cross-verified by reports from multiple media outlets, including Guangzhou Daily's Xinhua City, on March 9th, the day of the public beta launch. These individuals came from diverse backgrounds: HR, administration, operations, and sales—none of them relied on coding skills.

The starting point for this product line wasn't "we judged that office agents were a good market segment," but rather "a group of people who can't code are already making do with programmers' tools; let's give them something that doesn't require making do." From day one, WorkBuddy targeted non-technical users, not because of what market analysis said, but because user behavior already spoke for itself.

Three design decisions

In an interview with TMTpost, Wang Shengjie said, "Users don't need to understand what an agent, a tool, or prompt engineering is. They only need to know what they want."

This sentence summarizes WorkBuddy's first design decision: replacing technical concepts with natural language. In WorkBuddy, if a user inputs, "Help me organize last week's sales data and create a comparison table by region," the system automatically breaks it down into data retrieval, cleaning, analysis, tabulation, and output. Users won't see terms like "Agent scheduling," "tool invocation," or "context management" on the interface. In contrast, Codex and Claude Code's interaction starts with terminal commands or IDE plugins, requiring them to understand model behavior, manage token budgets, and handle execution errors—interaction logic designed for engineers, not HR.

The second decision was to pre-package scenario templates. The Beijing News, in its public beta report, revealed that WorkBuddy has over 20 built-in Skills packages, covering data processing, invoice processing, document archiving, competitor research, content creation, public opinion analysis, and sales insights. Each Skill is backed by a pre-set workflow that users can use immediately without needing to design automated processes from scratch.

In contrast, OpenAI didn't launch its role plugins in Codex until June 2, 2026, with an initial six covering data analysis, creative production, sales, product design, public equity investment, and investment banking. Anthropic's Claude Cowork takes a different approach. Instead of providing pre-built templates, it allows users to directly manipulate local files and applications on their desktop using natural language. This approach stems from an embarrassing discovery: Anthropic's official product page admits that the company's internal marketing and data teams are bypassing the chat interface and directly using Claude Code, which is geared towards developers.

The third decision was to make it native to the ecosystem, rather than through add-ons. A report by TMTpost points out that WorkBuddy's integration with Tencent Docs isn't via API calls, but rather a seamless integration. Users can directly invoke the Agent within Tencent Docs to process the current file, eliminating the need to switch between the two applications. It also supports direct remote control via WeChat; tasks initiated on the computer can have their progress viewed and additional instructions added on the phone before switching back to the computer to continue. For domestic enterprises that already heavily utilize WeChat, WeChat Work, and Tencent Docs, WorkBuddy isn't "just another AI tool," but rather an added feature to their existing workflows.

Codex takes a different approach. OpenAI's June 2nd announcement listed 62 integrated applications, ranging from GitHub and GitLab to Salesforce, HubSpot, and Snowflake. This is a general integration strategy targeting the global market, offering sufficient breadth, but the depth depends on the quality of each third-party API. Claude Cowork, on the other hand, focuses on local file systems and desktop application operations, with less emphasis on native integration with office suites.

The three decisions correspond to the removal of three barriers: cognitive barrier (no need to understand technical concepts); scenario barrier (no need to break down task processes yourself); and environmental barrier (no need to leave the office software you are currently using). Codex's role plugin and Claude Cowork are both moving in the same direction, but WorkBuddy delivered a complete solution for non-technical users ahead of schedule in the first half of 2026.

What do the growth data say?

Changes in user demographics are ultimately reflected in the numbers.

Analysys's "China Office Intelligent Agent Platform Market Research Report 2026" provides a clear ranking: as of May 2026, WorkBuddy's monthly PC visits reached 8.85 million, leading the second-place competitor by 2.6 times, with a month-on-month growth rate of 72.2%. Tencent's Q1 2026 financial report confirms this from another perspective. DoNews, in its financial report interpretation, quoted Tencent as saying that, in terms of daily active accounts, WorkBuddy has become the most popular efficiency AI intelligent agent service in China. According to Touzhong.com, WorkBuddy's month-on-month growth rate reached 831% in March. During the same period, Titanium Media reported that on the day of its public beta launch, the number of requests far exceeded that of CodeBuddy, triggering an emergency expansion due to computing power alarms, ultimately resulting in a 10-fold increase in capacity.

The increase in the proportion of non-technical users is currently only described qualitatively; the phrase "significantly improved" appeared in OmniTools' June 16th newsletter, but specific percentages are not publicly available. If such percentages were available, they could be directly compared to Codex's 20% non-developer ratio. However, the accuracy of user profiling on the WorkBuddy side is still weaker than that on the Codex side.

Even so, the direction is clear. The growth curve and the timing of the influx of non-technical users highly coincide, and the growth rate after the public beta in March, along with the traffic entry effect of the Tencent ecosystem, have jointly propelled WorkBuddy to the top position.

Overseas products are turning around

Both OpenAI and Anthropic have seen the same signal.

The data disclosed in OpenAI's June 2nd announcement itself is a signal: the growth rate of non-developers is more than three times that of developers. This is not just the growth of a peripheral user group, but the formation of a new driving force for growth. The six role plugins and Codex Sites feature released on the same day represent a clear "non-technical shift."

Anthropic's story is more straightforward. The first paragraph of Claude Cowork's official product page acknowledges the product's origins: the company's non-technical team first "voted with their feet" by using Claude Code, and only then did Cowork come about. Anthropic's conclusion is that "most AI assistants require users to break down tasks into individual prompts, while Claude Cowork accepts a result and handles the rest itself." Forte Labs, in a comparative analysis, points out that Claude Code requires command-line installation and is geared towards developers. Claude Cowork offers a simplified interface, but its positioning remains as a standalone desktop operation for handling local files and applications, not designed for scenarios like "directly editing reports in Tencent Docs."

Both products are changing course. But for tools that originated from CLI or IDE, transforming the interaction paradigm from command line to GUI and expanding the user model from developers to HR and administration requires more than just adding features; it also involves restructuring the permission system, governance framework, scenario templates, and ecosystem integration methods. This transition itself takes time. The window of opportunity WorkBuddy gained in the first half of 2026 is precisely the gap in this transition cycle.

Two paths, not one is better than the other.

WorkBuddy's approach is to build an agent within users' office software. WeChat, Tencent Docs, Enterprise WeChat, and QQ are tools users already use. The agent isn't the destination, but a capability layer embedded in the existing workflow. Codex and Claude Code, on the other hand, choose to build an agent and then let users find it. Codex is in the terminal, in the IDE, and in the ChatGPT application. Claude Code is even only in the terminal, while Claude Cowork is on the desktop. These products have stronger core capabilities and more flexible models, allowing developers to build workflows in any scenario. However, for non-technical users to "reach" this agent, they must first overcome the hurdles of installation, configuration, and understanding the interaction logic.

Wang Shengjie's assessment: "Users don't need to understand what an Agent is; they only need to know what they want." The difference between this product philosophy and the product philosophies of Codex and Claude Code lies in the fundamental divergence between the two approaches: "scenario encapsulation" and "underlying capabilities." The former pursues "the shortest path for users to find the Agent," while the latter pursues "the most things the Agent can do for users."

In the domestic market, WeChat and WeChat Work cover a much larger user base than mobile devices, and WorkBuddy's starting point is inherently closer to non-technical users. In overseas markets, GitHub and mobile devices are the domain of developers, making it perfectly reasonable for Codex and Claude Code to start from there. The influx of non-technical users is unexpected growth, and it will take time to catch up.

Pricing reflects this difference as well. WorkBuddy's personal Lite plan is priced at 39 RMB/month, Standard at 99 RMB/month, and Pro at 299 RMB/month. The enterprise SaaS flagship plan is 198 RMB/person/month, a pricing structure confirmed by Tencent Cloud's official pricing announcement and third-party channels. Codex requires a ChatGPT Plus subscription to start at $20/month, as shown on OpenAI's official pricing page. Claude Cowork also starts at $20/month with its Pro plan, as seen on Claude's official pricing page. WorkBuddy's lower starting price naturally gives it a wider reach in the Chinese market.

How long is the leading window?

WorkBuddy was about six months ahead in "scenario encapsulation," but the pursuers were not far behind.

In its June 2nd announcement, OpenAI outlined a roadmap for over 11 role-based plugins, with non-technical users growing at more than three times the rate of developers. Anthropic positions Claude Cowork "for non-technical tasks," a strategic product shift. Neither team considers office agents a side business.

WorkBuddy's enterprise version will be released on June 5, 2026, including the "Projects" feature, an enterprise management backend, and 24/7 expert digital staff. The enterprise version is a crucial step in maintaining its advantage, expanding from individual users to organizational-level procurement, with completely different pricing models and retention strategies. Codex and Claude Code's accumulated expertise in underlying model capabilities is undeniable, while OpenAI and Anthropic possess the world's most advanced model training capabilities. Once they complete their transformation of the interaction layer, WorkBuddy's first-mover advantage will face more direct competition.

A 46-year-old government cloud salesperson was able to generate a city economic analysis report on WorkBuddy in 15 minutes, a case reported by TMTpost. An independent developer created six digital jobs, replacing 50,000 to 80,000 yuan in labor costs with a monthly cost of 800 yuan. When non-technical users can enable agents to complete tasks independently without learning anything, the difference in daily active users is not an outlier, but a natural result of differences in skill levels.