Tencent launches a full-scale "Lobster Counterattack"
```
Authors: Song He, Huang Yu
On March 11, Tencent announced the launch of the Skills community—SkillHub—optimized for Chinese users, an AI platform based on the OpenClaw (Lobster) official open source ecosystem, providing localized configuration services and compatible with skills from its official community.
This is just another footnote in Tencent's wide-ranging counterattack in AI around the OpenClaw ecosystem over the short span of six days.
Starting from 10 a.m. on March 6, 2026, Tencent arranged engineers at the North Plaza of its headquarters in Nanshan District, Shenzhen, to offer free OpenClaw installation services. Visitors lined up in a queue that stretched to the opposite side of the street.
Even more surprising, at the Guangdong delegation's national legislative group meeting the next day, Gao Wen, academician at the Chinese Academy of Engineering and director of Pengcheng Laboratory, made a comment that brought a knowing smile to the room—"The lobster craze is so hot, even Pony Ma didn't foresee it."
Perhaps Pony Ma truly hadn't expected it. But Tencent’s reaction speed has indeed exceeded general market expectations.
A mere three days later, on March 9, Tencent rolled out three Lobster products at once: The self-developed full-scenario desktop agent WorkBuddy—zero deployment, zero configuration, ready to use in browser; the quietly launched internal beta QClaw, a local AI assistant based on OpenClaw—one-click install for both Windows and Mac, supporting remote control of the local PC and full data localization via WeChat; and an official plan for full OpenClaw integration with the QQ Open Platform.
On March 10, Tencent's official account gave these products a visually striking collective name—“Lobster Task Force.”
If you also include previously launched Tencent Cloud Lighthouse cloud solutions, with OpenClaw users exceeding 100,000, the enterprise-grade agent development platform ADP, Tencent Computer Manager’s AI security sandbox “Lobster Butler,” the cloud-based AI Agent Security Center “Cloud Security,” and the lobster-integrated version of Tencent Lexiang Knowledge Base...
From individuals to enterprises, from local to cloud, from security to knowledge—within less than two months, Tencent has forcibly pieced together a full-scenario Agent product matrix.
As Pony Ma put it in his WeChat Moments: “Self-developed lobsters, local lobsters, cloud lobsters, enterprise lobsters, cloud desktop lobsters, security-isolated lobster rooms, cloud security, knowledge base... and a batch of products are coming soon.”
This is a counteroffensive with no prior announcement, and what makes it dramatic is that just a few months ago, the market was still mocking Tencent as an “old stock” in the AI age.
But everything happening now is far from the endpoint of Tencent's AI strategic deployment.
From the current Lobster matrix to the future WeChat Agent, in a sense, this 5 trillion Internet giant’s counterattack has just begun.
Giants’ Decisive Moment
To understand Tencent's seemingly counterintuitive full entry into Agents, we may first need to look at how it lagged at the start of the AI age.
In May 2023, amid the fever of the “Hundred Large Models Battle,” Pony Ma made a remark at the shareholders’ meeting that surprised the market: Tencent is not in a hurry to quickly finish up and show off half-finished products. For the Industrial Revolution, launching the light bulb one month early isn’t all that important in the long run.
Though this sounded strategic, the market was not convinced.
In 2024, Doubao swiftly captured the 2C AI application market through traffic skews on Douyin’s feed and fast product experience, with DAU surpassing 100 million by the second half of the following year. Early 2026 saw Alibaba’s Qianwen aiming to win user mindshare in “helping users get things done,” with full integration across Alibaba’s ecosystem—Taobao, Alipay, Fliggy, Amap, etc.—in January. Baidu Smart Cloud also embraced the OpenClaw open-source ecosystem, launching one-click deployment on February 3, just before Spring Festival.
By contrast, Tencent's Hunyuan large model lagged behind ByteDance and Alibaba at the C-end, and attempts to promote Yuanbao through viral spread on WeChat’s social chain were often seen as overly intrusive to users.
Even with 1 billion yuan of red packet marketing during Spring Festival pushing daily active users to over 50 million, there were still doubts about whether paid user acquisition could buy loyalty.
A person close to Tencent told Wallstreetcn that compared to large language models, Tencent invested more energy into multimodal large model development, which led to Hunyuan LLM not making it into the industry’s top tier.
Thus, a harsh comparison emerges: Tencent, sitting on two super entry points—WeChat and QQ—with a combined monthly active user base of over 1.5 billion, was beaten by ByteDance and Alibaba in AI applications.
Investors began to worry. After Tencent’s share price peaked at HKD 683 in October 2025, it began a months-long stepwise decline, finally falling to 560 HKD in early 2026, with market cap once dropping below 5 trillion HKD.
At the same time, social media began labeling Tencent as a “Tech Old Stock”—implying its fundamentals were solid but innovation was slow, resembling a “retired old man” in the AI age, living off past glories.
Pony Ma was not oblivious to the problem.
At a rare employee meeting on January 26, 2026, he frankly addressed criticism about being “too slow.” He admitted that various business units’ priorities to integrate AI with their own scenarios led to overall sluggishness, and only by the end of 2024 did the Yuanbao team complete its integration from TEG to CSIG. He also announced a restructuring of the AI R&D team—hiring former OpenAI senior researcher Yao Shunyu as Chief AI Scientist in December 2025, and establishing the AI Infra, AI Data, and Data Computing Platform divisions.
Yet he added a thoughtful remark: “Every enterprise has different genes and constitutions. Tencent’s style is steady and solid.” He emphasized that AI is currently the only area truly worth heavy investment for Tencent. Tencent’s AI strategy has been “self-development + open source”—“similar to our previous development in gaming, always balancing in-house and agency games, leading to the best user experiences.”
Looking back, this statement had already planted a seed.
Tencent has never shied away from open source at the model layer—first fully accessing DeepSeek-R1, sending Yuanbao’s DAU skyrocketing more than 20-fold into China’s top three; then, as OpenClaw went viral, Tencent Cloud Lighthouse became the first to launch a one-click deployment template on January 28, even before Alibaba and Baidu cloud.
The so-called “being half a step slow” may just mean Tencent was waiting for a battlefield where it truly excels.
The OpenClaw explosion contains an easily overlooked detail: The framework first appeared at the end of 2025 as Clawdbot, with little attention at first. The real market trigger came after the release of Claude Opus 4.5—same framework, same code, but results only worked after switching to a stronger underlying model.
Bear in mind: In 2024, there were several similar Agent projects but none succeeded; this time, the agent is finally supported by the model.
The entry threshold for OpenClaw is no secret: configuration is too complex for ordinary people to install. Setting up Node.js, troubleshooting errors, opening Feishu or Qiye WeChat API permissions—each step is a challenge for non-technical users.
This is precisely Tencent’s best battlefield.
Looking back at Tencent’s twenty-year product history, a consistent logic emerges: It may not be a technological originator, but it almost always leads in “turning complex technologies into products accessible to everyone.” QQ, WeChat, WeChat Pay, mini-programs—all tell this story.
On the Lobster chessboard, Tencent is again taking this path—not as a creator of foundational models, but as the biggest Agent-era product platform.
Volcano Engine CEO Tan Dai once predicted: “The PC era was Web, the mobile era was apps, the AI era is Agents.” If so, Tencent’s task is to be the “App Store” and “WeChat” of the Agent era.
When facing doubts, Pony Ma, at the 2026 employee assembly, gave certainty to staff from the outset: stay steady and focus on your own pace. He said: “Every enterprise has different genes and constitutions. Tencent’s style is steady and solid.”
Clearly, after three years of “hibernation,” 2026 will see Tencent speed up its AI strategy. An internal source revealed, “We expect to see more big moves by Tencent in AI this year.”
The Future of WeChat Agents
As outsiders have noted, Tencent has developed a precise and layered “Lobster Task Force” product matrix around OpenClaw this March, targeting not only novice users but also hardcore developers and large enterprises, aiming to smash down usage barriers.
But while everyone’s eyes are on the Lobster product matrix, the real trump card remains hidden below the surface.
At the 2025 annual results meeting, responding to analysts on the Agent strategy, Pony Ma said something crucial: “Anyone can build a general agent, but meanwhile, there are agents that can exist within WeChat and its unique ecosystem. I think these are two different products.”
He further explained: For generic agents, Tencent is building these in Yuanbao and other AI-native products, which may start by quickly answering questions, then gradually add chain-of-thought and long-chain reasoning, and after that enable more complex automation—“But these are not too different from what our peers provide.”
The thing that truly lights up Pony Ma’s eyes may be the internal WeChat agent.
The WeChat ecosystem has unique components—social graph, communication and community capabilities, content ecosystem (including Official Accounts, Channels), and millions of mini-programs within WeChat.
“These mini-programs involve various information and transactional/operational capabilities across many verticals. So, compared to other general agents, this will be very unique. For Tencent, it’s a highly differentiated product,” Pony Ma once said.
This remark is significant. It implies that Tencent’s top management has explicitly positioned the WeChat Agent as a strategy-level product independent of a simple general AI assistant—not just “adding an AI feature to WeChat.”
But agentizing WeChat is much more complex than outsiders may imagine.
At the January 2026 staff meeting, Pony Ma appeared cautious about making WeChat “AI-ized”: “An all-in-one AI package is not necessarily everyone’s favorite. In the future, we’ll stick to a decentralized approach to balance user needs and privacy/security in WeChat’s smart ecosystem.” He particularly noted that the WeChat business group should “be planned carefully and move only when ready.”
This “cautiousness” has deep logic behind it.
WeChat’s 1.3 billion MAU is China’s scarcest Internet asset, covering almost everyone’s social relationships, payment behavior, information consumption, and life services.
A wrong product decision could cause irreversible user trust loss. The “restraint” philosophy established under Allen Zhang—resisting excessive commercialization and pushy features—was key to WeChat defeating IM competitors from the QQ era and dominating mobile Internet.
But the Agent era fundamentally challenges this philosophy.
An agent’s core ability is “acting for you”—it needs access to your contacts, to read your messages, access your payment account, and operate your collection of mini-programs. This naturally clashes with WeChat's longstanding privacy-first approach.
This likely draws a red line for Tencent’s own agent path—WeChat’s agentization must be advanced within a secure and controllable framework.
Thus, stepwise piloting before full rollout may be the path Tencent must take at present.
Tencent has a time-tested product logic: new features and formats are first validated in QQ, then gradually rolled out to WeChat once mature.
The deployment rhythm of Lobster precisely echoes this pattern.
Step one is full rollout on QQ. On March 7, QQ Open Platform officially let individuals access OpenClaw, enabling AI chatbot creation with zero coding. QQ’s young user base is open to new tech and willing to experiment, making them natural early Agent users. Compared to Feishu and DingTalk, implementation on QQ is easier.
Step two is QClaw’s detoured approach. The WeChat Open Platform has never provided an official bot API for personal WeChat.
All current personal WeChat bot solutions in the market are “black box” workarounds using protocol reverse engineering and hooks, risking account bans and compliance issues. QClaw’s ingenuity: it connects to personal WeChat through the Enterprise WeChat "WeChat Customer Service" entry—Lobster runs on the user’s own computer, with WeChat merely issuing commands.
This is a cloud/local separation compromise: users can converse with Lobster in WeChat, without requiring WeChat to open up core APIs.
If, after full launch, QClaw also supports direct personal WeChat connection, this would be WeChat’s first ever official Agent bot integration.
Step three is the ultimate goal: native WeChat lobster agent. Although not officially confirmed, multiple reports say Tencent is planning an Agent embedded directly in WeChat chat boxes, linking millions of mini-programs to handle daily tasks like ride-hailing, ordering food, registration, or booking tickets, possibly launching in 2026 Q2~Q3.
Challenges Remain
The nascent WeChat Agent still faces challenges.
First, its main obstacle is at the frontend; its biggest hidden opponent is WeChat itself.
WeChat has long been famous for “restraint,” as Allen Zhang repeatedly expressed wariness of excessive functionality.
Every major iteration—from Moments to WeChat Pay to mini-programs to Channels—has gone through lengthy internal debate and careful gray testing.
This culture secures product quality and user trust, but means WeChat naturally moves slower than competitors in new tech waves.
Pony Ma’s comment at the staff meeting: "How to leverage one’s own advantages to upgrade with new technologies—give yourself more time, plan thoroughly before moving."
This both calms external anxieties and protects the WeChat team. But WeChat’s market window is not infinite.
If WeChat Agent can’t land soon, users may get used to using Doubao or Qianwen for “getting things done.” Once habits form, switching—even to a better WeChat Agent—becomes difficult.
Tencent may use QClaw-type clusters as “unofficial pilots” to measure genuine user demand. The behavioral data captured by these products might directly influence when and how WeChat officially introduces Agents.
To some extent, QClaw also acts as a real-time “feasibility survey” for WeChat Agent.
Second, backend model competence is the short board in urgent need of improvement.
Though Tencent’s self-developed Hunyuan model leads in multimodal, for general language ability, Doubao, Qianwen, and DeepSeek have captured more user mindshare first.
Tencent now takes a “self-developed + open source” dual-core strategy. Tong Daosheng at the annual results meeting underscored: Core technology will be in-house, but open source models are also embraced, letting users select best-fit models for different scenarios. WorkBuddy supports domestic models like Kimi, MiniMax; QClaw can integrate custom models.
The wisdom of this approach is “not putting all eggs in one basket”—not demanding in-house models win out in all dimensions, but providing users with the “best model routing” via product-level integration.
Yet the risk is clear: If Tencent never masters a strong enough general-purpose LLM, it is always the “assembler,” not the “core supplier”—profits from APIs and Tokens get split by model providers.
Pony Ma stated in the 2025 Q3 report: “We are continuously improving the Hunyuan foundational team and architecture. As Hunyuan grows, our Yuanbao push and AI agent development in WeChat will make better progress.”
In all of 2024, Tencent's R&D investment reached 70.69 billion yuan, with 61.98 billion in the first three quarters of 2025, a 22% YoY increase. Liu Chiping revealed that 2025 capex will be a “low double-digit percent” of total revenue—meaning near the 100 billion mark.
Money is being thrown at the problem, but model advancement takes time. Whether Yao Shunyu—the former OpenAI chief scientist hired by Tencent—can achieve a leap in Hunyuan’s general abilities within 12–18 months is a key variable for Tencent’s Agent strategy success.
Third, how to balance agent costs and WeChat’s commercialization is a long-term question Tencent must answer.
As is well known, running a complex OpenClaw task can trigger thousands of model calls. A Guolian Minsheng Securities report likened tokens to the “energy of the Agent era”—“raising lobsters” is not a one-off cost, but an ongoing one.
If Tencent is to empower all 1.3 billion WeChat users with Agents, token costs will be astronomical.
Short term, offsetting costs with advertising earnings or business growth “subsidizing” AI investment is likely Tencent’s practical plan for now. Pony Ma said on the Q3 earnings call: “AI is not a cost; it’s the engine for all our business growth.”
In Q3 2025, AI-driven ad targeting pushed marketing services revenue up 21% YoY to 36.2 billion yuan, keeping double-digit growth for 12 quarters. AI makes the “money printer” of ads run more efficiently—this may, short term, be the main funding source for Agent investment.
But long term, Agent business models must still cover the massive, unavoidable token expenses behind the scenes.
A source close to Tencent analyzed three ways to commercialize costs: 1. Token consumption fees—direct charges for cloud deployment, but individual user fees are unlikely; 2. Mini-program transaction commissions: for every user transaction in mini-programs via the Agent, WeChat can charge a channel fee; 3. Enterprise SaaS subscriptions: the ADP platform charges corporate clients. For example, OpenClaw users on Tencent Cloud Lighthouse topped 100,000 and keep rising, while developer usage repeatedly hits record highs—the “selling shovels” cloud model is taking shape.
The Endgame
Perhaps all current moves need to be reconsidered from the first-principle question:
What does this massive Lobster counterattack really mean for Tencent?
On the surface, it is a giant quickly following the open-source Agent craze. But from a longer perspective, it might be the starting point for Tencent’s transformation from “social platform company” to “intelligent infrastructure company.”
In the PC Internet era, the entry point was the browser and search engine. In the mobile Internet era, it was apps and app stores. In the Agent era, it will be the agents embedded in daily communication tools—no need to open any apps, just say it in chat, and AI will handle everything.
Whoever holds the largest chat box will have the ultimate entry point of the Agent era.
In China, that chat box is indisputably WeChat.
As Pony Ma said at a results meeting: “AI is still in early development. Just like when the second industrial revolution arrived—with the power plant, power grid, and the light bulb as the earliest applications, later appliances and the Internet appeared thanks to that ecosystem.” His meaning: we are still building light bulbs.
But Deloitte predicts Agentic AI is growing at a 43% annual rate, outpacing other AI tracks. CB Insights reports that, since 2023, Agent mentions on earnings calls have increased 10-fold. Silicon Valley VC godfather Vinod Khosla predicts 2026 will be the year agents truly mature and show visible impact.
Maybe the light bulb era is ending; when the age of appliances arrives, the ones holding the biggest power plant and densest grid will win.
From entry point to closed loop, countless obstacles remain. Can models execute complex tasks reliably? Will the security system reassure users to release control? Can WeChat’s ultra-cautious openness culture make way for agents?
These questions await Tencent’s stepwise probing, attacks, and constant user feedback for answers.
But something already changed.
In early March 2026, as people flocked from Hangzhou and Hong Kong, queuing for Tencent engineers to install that little red lobster, a clear signal appeared: AI is no longer just a geek’s toy or an investor’s concept. It may be flowing into every ordinary person’s computer, phone, and life.
Tencent may now be standing at the estuary of this river.
Risk Warning and DisclaimerMarkets are risky and investments must be made cautiously. This article does not constitute personal investment advice and does not take into account users’ specific investment goals, financial situations, or needs. Users should consider whether the opinions, viewpoints, or conclusions in this article fit their particular circumstances, and bear responsibility for investment decisions accordingly. ```