From the sudden emergence of OpenClaw earlier this year, to tech enthusiasts sharing their "lobster-raising" experiences on social media, to ordinary people queuing up at Tencent's headquarters for a free installation of OpenClaw, the spring of 2026 saw a "lobster storm" triggered by desktop agents quietly reshaping the work habits of office workers.
Amid this surge of efficiency revolution, Tencent's WorkBuddy officially launched on March 9. User traffic surged dramatically, leading to "growing pains" for WorkBuddy on its second day, as core services were overwhelmed, forcing an emergency tenfold expansion to handle the market's enthusiasm.
This scenario of immediate service strain upon launch highlights the high level of interest from non-developers in "desktop agent-type products."
As the product lead for Tencent's WorkBuddy and head of Tencent Cloud's developer products, Wang Shengjie witnessed the entire journey of WorkBuddy from incubation to explosive popularity.
On March 13, during an exchange with media outlets including Wall Street Insights, Wang Shengjie revealed that WorkBuddy's market performance post-launch far exceeded the team's expectations. Instant requests surpassed those for CodeBuddy by many multiples, and computing power was insufficient, hitting warning thresholds. "We urgently expanded capacity and optimized the architecture to make the login logic more stable," he stated.
WorkBuddy was developed by the Tencent Cloud CodeBuddy team. CodeBuddy is an AI programming tool released by Tencent last year, while WorkBuddy is built on the same Agent architecture, forming a desktop agent platform with a foundation, an ecosystem, and scalability.
The birth of WorkBuddy was not a hasty attempt to follow a trend, but rather a long-planned "blitzkrieg." As early as the second half of 2025, Wang Shengjie's team had already built the underlying architecture and open platform for CodeBuddy, which supports AI autonomous task execution, and released an SDK. This approach of first solidifying the execution infrastructure and then opening the ecosystem aligns completely with the core logic behind Anthropic's creation of Claude Cowork.
Wang Shengjie recalled, "After New Year's Day this year, when Claude Cowork was released, I took the idea to my boss. Based on our own platform, we quickly iterated to create a prototype of WorkBuddy. My boss immediately agreed, thought it was excellent, and we got to work."
Wang Shengjie specifically noted that before OpenClaw's viral success, the strongest product in this sector was actually Claude Cowork, but it failed to gain widespread traction because it was an overseas product and not open-source.
After just two all-nighters, version 0.01 of WorkBuddy was launched, though initially it was primarily for internal use at Tencent. "I still remember we started on January 17th, which fell on a Saturday and Sunday. My colleagues and I pulled two all-nighters," he said.
The outside world first learned that Tencent was developing a desktop agent tool similar to OpenClaw on February 6th. On that day, the official account of "Tencent Cloud Code Assistant CodeBuddy" announced the start of WorkBuddy's internal testing.
Following the viral success of OpenClaw after the Spring Festival holiday, the CodeBuddy team accelerated the process for WorkBuddy's official launch.
Wang Shengjie expressed that the key to OpenClaw's sudden popularity lies in its combination of open-source capability and its ability to meet automation needs in office scenarios. The mission of WorkBuddy and OpenClaw is very similar, but their implementation methods differ.
Simply put, WorkBuddy follows a path of self-developed productization, security, controllability, and ecosystem integration, while OpenClaw follows a path of open-source openness, high freedom, and community-driven development.
Wang Shengjie stated, "When we incubated WorkBuddy internally, we focused more on the user experience and security, helping users solve their problems within a specific, controllable scope."
He emphasized that WorkBuddy is 100% self-developed and does not use a single line of OpenClaw's source code. "Because we already have the entire infrastructure of CodeBuddy—the framework for AI-driven, autonomous task completion—we didn't need to borrow anything. We started from user scenarios and created a product that aligns with the form of OpenClaw; it's not just a shell."
As the focus of competition in the AI industry shifts from "comparing model parameters" to "comparing implementation capabilities," agent products are booming. Now, OpenClaw seems to have become a synonym for this type of product, and the launch of any similar product is easily seen as following OpenClaw's trend.
Concurrently, Wang Shengjie explained that the "lobster" is a concept with different implementation ideas. Currently, WorkBuddy is not a purely托管 automation product; in terms of automation level, it can be considered to be in a middle position.
The WorkBuddy team's assessment is that what most users truly need now are scenarios involving search plus content reorganization. Examples include daily AI news analysis or automatically converting Twitter content into a Xiaohongshu style and posting it.
"This is already a quite advanced scenario, and it's safe," Wang Shengjie noted. In his view, there is no need for AI to autonomously post or learn without user authorization, calling that prospect "quite frightening."
"Some overseas products are attempting higher levels of autonomy, even full托管 direction. Only after everyone truly understands which scenarios hold value will WorkBuddy consider布局 full automation," he added.
OpenClaw has thoroughly popularized the "lobster" concept, transforming it from a specific product name into a productivity concept representing "autonomy and remote operation." However, competition after the initial红利 fades often comes with scrutiny over security and boundaries.
In this unpredictable battle for technological survival, the player who can truly understand users' pain points will be the one to raise the strongest "lobster" in the jungle of computing power accumulation.
Comments