The desktop Agent of TENCENT Cloud is a form of subtraction

Author | Huang Yu

The major internet companies are gathering in another hot new battlefield—desktop Agent.

On February 6th, when the official account of "Tencent Cloud Code Assistant CodeBuddy" announced the launch of internal testing for the desktop Agent tool "WorkBuddy," this highly anticipated "desktop exam" welcomed another heavyweight player.

The Tencent Cloud CodeBuddy team told Wall Street Insight that WorkBuddy was launched internally at Tencent quite early, aiming to help non-developers within the company improve their daily efficiency. Now, seeing the increasing demand from many office workers, researchers, and various industries for "applications to solve daily tasks," they have launched a lightweight and minimalist product design style based on the currently mature AI Coding infrastructure of CodeBuddy.

"We will continue to refine based on user feedback and find a suitable time to officially launch this product in the first half of the year."

Unlike the currently popular OpenClaw, which can access system-level permissions and operate more flexibly, WorkBuddy is positioned more as "performing various autonomous operations by reading authorized folders on the computer."

This means that the scenarios for WorkBuddy are more about limited operations under clear folder constraints.

This may also reflect Tencent's "subtractive" product thinking in this field— it assumes that most office workers not only care about "whether they can get the job done" but also worry about whether AI will cause problems, so the chosen product route is not aggressive but emphasizes stability, reusability, and deliverable execution results.

The launch of WorkBuddy not only marks Tencent's official entry into the market competition for desktop Agent tools but can also be seen as a significant signal that major companies are trying to redefine the boundaries of workplace efficiency through this new variable of "intelligent agents."

Tencent's timing for entering the market coincides with the critical point of the global desktop Agent explosion.

Just recently, products like OpenClaw have rapidly gained popularity in the tech community, with users amazed at how AI can deeply access computer systems, files, and even autonomously execute cross-application tasks.

This experience, referred to as "24-hour standby Jarvis," has led the industry to declare that the "ChatGPT moment for desktop Agents" has arrived.

Compared to the currently popular similar tools on the market, WorkBuddy does not take a purely geeky route but is positioned for office workers without a technical background.

This means it no longer requires users to have programming knowledge; they only need to describe their needs in natural language, and it can execute tasks like a "knowledgeable colleague." This shift from "technical tool" to "workplace assistant" is Tencent's bold attempt at popularizing desktop Agents.

According to reports, WorkBuddy can autonomously plan and execute multimodal complex tasks directly on the local computer. Specifically, WorkBuddy has two major features: First, deep local operation capability: it can "understand human language" and "truly operate local files."

After obtaining authorization, it can read folders, batch process files, generate documents or PPT data deep analysis, industry research, and multi-task Agent parallel processing, etc. This means it breaks the awkward situation of traditional AI "only being able to see, but not act," truly addressing the last mile of office work.

Second, the autonomous planning of multi-step tasks, unlike simple instruction responses, WorkBuddy emphasizes "thinking with a brain."

In addition, WorkBuddy also has built-in various domestic and international models, mainstream MCP Server, and Skills skill packages, among other advanced features, which allows it to maintain simplicity and ease of use while possessing strong scalability, capable of meeting the deep needs of vertical segments such as finance, research, and creation.

Wall Street News learned that WorkBuddy was developed by the Tencent Cloud CodeBuddy team, which has previously launched it internally at Tencent and started internal testing. Currently, more than 2,000 employees from different positions are participating in deep usage, and it has been widely applied in scenarios such as data processing and analysis, building local knowledge bases, content copywriting, poster generation, and automated office work.

For a long time, although large models perform omnisciently in dialogue boxes, when users return to real office scenarios, facing complex local files and trivial multi-task switching, large models often appear "handicapped."

Now, as AI competition enters the second half, the focus of industry competition is shifting from "comparing model parameters" to "comparing implementation capabilities." A simple chatbot can no longer meet the complex, fragmented, and highly localized file-dependent needs of workplace individuals. Whoever can first occupy the user's computer desktop will gain the ticket to the next generation of "AI operating systems."

The popularity of tools like OpenClaw has already proven that users have an extreme thirst for Agents that "can run on their own computers and deeply operate applications."

Tencent's move at this time is not only to seize an ecological position amid a host of pioneers but also to upgrade its long-cultivated enterprise service capabilities through the Agent medium.

However, behind the rekindled market sentiment, WorkBuddy and a host of desktop Agent players also face challenges.

The foremost challenge is the game of "overreach" and "privacy." The core value of desktop Agents lies in their deep access to users' computer systems, files, and applications, which means users must grant extremely high system permissions. Just as the industry expects OpenClaw to be a "24-hour standby Jarvis," the more powerful the Agent, the more detailed the personal preferences and local project data it holds.

How to ensure that these sensitive data are not misused or leaked during network transmission while providing convenience will be a compliance "life-and-death line" that AI Agent products must face.

The Tencent Cloud CodeBuddy team told Wall Street News that WorkBuddy relies on CodeBuddy's security protection mechanism and employs a multi-layer defense strategy of "sandbox isolation + Skill standardization + dangerous operation interception" to address the deployment challenges of desktop Agents Its core idea is: "Authorize trust and verify" — trust the user's command intent, and verify the execution safety and dangerous operation blacklist through user authorization.

In addition, the "parent-child relationship" between the Agent and the underlying model also determines its upper and lower limits of capability.

Although desktop Agents can optimize execution paths through engineering means, the fundamental capabilities of the Agent are still exhibited by the underlying large model. Once the underlying model experiences "hallucinations" or logical breaks, the Agent may also bring serious consequences in automatically executing tasks.

Moreover, it is worth mentioning that the rise of desktop Agents is essentially reconstructing the interaction relationship between humans and computers. In the past, users acted as "movers" between applications, shuttling back and forth between browsers, office suites, and communication tools; however, with the popularity of desktop Agents like WorkBuddy, the future interaction paradigm may shift to "task-centric."

When an Agent can cross applications and represent users to complete complex tasks such as bulk unsubscribing from emails or organizing price discrepancies, traditional application software may gradually become "skill packages" called upon by the Agent. This means that in the future desktop ecosystem competition, it may no longer be about the accumulation of software functions, but rather whose interface is more conducive to Agent invocation and who can provide a more AI-compatible operating environment.

Globally, the competition for desktop Agents has just begun. Not only Manus, OpenClaw, and Claude Cowork, but also many other similar products are rushing to launch.

When AI truly takes over our mouse and keyboard, it represents the ultimate enhancement of efficiency and a new test of digital security boundaries. As the "desktop Agent war" intensifies, a transformation in productivity paradigms is already at our doorstep