WorkBuddy has exploded, and WeChat AI has arrived—has Tencent AI found a "new ticket"?
At Tencent’s shareholder meeting in May this year, Pony Ma said something that set a modest starting point for Tencent AI: “A year ago we thought we got on the boat, but later realized the boat was leaking. Now it feels like we're standing on it, but still can’t sit down. We still hope the boat can go faster.”
At that time, public opinion was widely discussing: Is Tencent AI slow? Yuanbao momentum was later than Doubao, Hunyuan release was behind Wenxin, AI products had insufficient presence.
But in just a few months, Tencent AI’s narrative seems to have completely reversed.
WorkBuddy is being discussed increasingly often, WeChat AI assistant “Xiaowei” has started internal beta testing, broker research reports are putting Tencent AI back in their valuation models. Over the weekend, investment communities even chanted “Tencent is great again”.
From “leaking boat” to red hot. Has Tencent AI found a “new ticket to board the boat”?
WorkBuddy Explodes
At the end of June, some unexpected data emerged.
Analysys data showed WorkBuddy’s monthly visits at 8.85 million, monthly active users (MAU) 20 million, daily active users (DAU) 13 million+, with the DAU/MAU ratio between 65-75%—this stickiness indicator is close to Slack-level office tools.
Tencent’s Q1 earnings report stated: “By daily active account, WorkBuddy has become China’s most popular productivity AI agent service.”
Even more notable is the speed: grew two orders of magnitude in 3 months, released 43 versions in 3 months, updated as usual during May Day holiday.
There are reports that Tencent internally positioned it as its “third phenomenon-level product” after QQ and WeChat. This appears to be Tencent’s “new ticket to board the boat” in the AI era.
This positioning naturally has some internal PR elements, but the numbers don’t lie.
WorkBuddy’s predecessor was CodeBuddy, an AI programming assistant for programmers. Development began in 2022, from limited internal testing at Tencent to covering 90% of the company’s engineers by the end of 2025, with most teams generating over 90% of their code by AI, and overall coding time reduced by 40%.
In January this year, the team expanded it to a desktop agent for general users, internal incubation of WorkBuddy version 0.01; officially launched in China on March 9; overseas version released on May 28.
Why did it take off? Two key points.
First, it removed the barrier of "installing lobster".
OpenClaw (“foreign lobster”) had too high a threshold—command line deployment, user-paid tokens, beginners were directly rejected, and this even spawned a business of “installing lobsters for others” on second-hand platforms.
WorkBuddy skipped all this: desktop client plus WeChat mini-app, minimal installation process, WeChat authorization allows sending instructions from mobile and PC will run tasks automatically. This is a capability no foreign “lobster” can replicate.
Second, Tencent’s ecosystem advantage is activated in the Agent era.
WorkBuddy has integrated Tencent Docs, Tencent Meeting, WeCom, Tencent Cloud Drive, TAPD, and through MCP protocol has connected to Feishu, Kingsoft, Qichacha, SalesEase, and 30+ external tools, expanding SkillHub skills library from 79,000 to 790,000.
Saying “turn last week’s sales data into a PPT and send to meeting colleagues”, it can fetch data, generate files, and send them itself.
In the previous “chatbot” era, Tencent’s ecosystem was fragmented; in the “help you get work done” Agent era, this tool chain becomes a moat.
As Liu Chiping said in Tencent’s Q1 earnings call: “Users can control AI agents within communication and browsing interfaces such as WeChat, QQ, and Tencent Browser.”
WeChat AI: The bigger card is just turning up
Besides WorkBuddy, an even bigger card is slowly being revealed.
On June 20, 2026, WeChat AI assistant “Xiaowei” started small-scale grey testing.
What can Xiaowei do? According to early feedback from internal test users: complete native WeChat operations via voice or text (send messages, Moments, make calls), use mini-programs to book appointments and order food, search and content generation, and generate mini-program prototypes via natural language.
This is a super app with 1.43 billion monthly active users, entering the agent track for the first time.
But we must be clear-eyed about its current status.
The JP Morgan research team (analyst Yao Cheng) in a report on July 6 provided a three-level value-creation framework, distinguishing what is already achievable from what needs to wait:
Level 1 (a): Assist with traffic funnel—agent helps users discover, compare, build orders in mini-programs, but checkout still requires manual user completion. Current public beta supports this scenario.
Level 1 (b): Closed-loop transaction—agent completes payment and transaction end-to-end. Not yet supported; AI wallet not integrated with main WeChat Pay wallet.
Level 2: Broad merchant integration, evolving WeChat from traffic source to half-marketplace platform. Probability increasing, but merchant adoption yet to be verified.
Level 3: WeChat becomes the dominant online consumer traffic entry, commanding a large share of merchant lead generation budget. JP Morgan assigns only 10% probability due to unresolved issues like potential cannibalization of ad business, recommendation algorithm neutrality, user trust, and regulation.
Weighted by probability, JP Morgan predicts WeChat AI agent could bring incremental revenue of RMB 126 billion by 2030 and incremental operating profit of RMB 88 billion, representing 18% upside to its 2030 operating profit forecast.
Goldman Sachs (analyst Ronald Keung’s team) is more cautious, saying in its June 23 report it has three core doubts:
First, Xiaowei uses WeChat’s self-developed WeLM model rather than the group’s Hunyuan, two parallel LLMs, unclear if resources are being duplicated;
Second, if widely rolled out, incremental inference costs could be 5%-17% of Tencent’s forecasted adjusted operating profit in Q4 2026;
Third, short-term monetization path is unclear; monetizing via ads needs AI to deeply penetrate local life, content discovery, and shopping scenarios.
Goldman’s conclusion: “Tencent’s multiple expansion in the next few quarters depends mainly on the progress of its AI narrative.” Its 12-month target price is HK$700 (SOTP method), about 62% upside from then, maintaining a buy rating.
The core disagreement between the two institutions is pacing: JP Morgan thinks the public beta turned WeChat AI from "abstract option" to "trackable milestone"; Goldman thinks short-term cost pressures and monetization mismatch require more quarter-by-quarter data validation.
How Competitors Responded
After WorkBuddy’s data came out, two immediate reactions:
ByteDance: On June 9, Trae upgraded from an AI coding tool to “Trae Work”, slogan changed to “Real AI Enabler”, added Work/Code dual modes, selling enterprise versions via Volcano Engine.
Alibaba: On July 2, merged QoderWork (office productivity), DingTalk “Wukong” (enterprise collaboration), Alibaba Cloud MuleRun (agent execution engine) under 1992-born Chen Yusen. Alibaba Partners Committee posted on internal network June 10 directly criticizing DingTalk’s management culture as “not what Alibaba should be”.
Quick leadership changes, mergers, culture critique—actions came fast, perhaps really triggered by WorkBuddy’s numbers.
The three companies’ paths differ:
- Tencent: “Agent + WeChat embedded”, government scenarios (2.3 million Guangdong civil servants already deployed) + WeCom as base;
- ByteDance: “Coding tool → office → enterprise version”, relying on Doubao/Seed model ecosystem;
- Alibaba: “DingTalk with 700 million user base + Agent Marketplace”.
Another noteworthy detail: WorkBuddy is the first phenomenon-level Tencent product in history to use ground promotion—QQ and WeChat never did this. Indicates Tencent’s determination to penetrate the B-side, not just talk.
Hunyuan Catch-Up: Tencent Finally Begins Model-Product Loop
In the past months, Tencent AI’s perceived slowness wasn’t just the application layer.
The underlying model also underwent reconstruction.
In official Tencent dialogues, Yao Shunyu said about Hunyuan 3 Preview: mainly rebuilt base infrastructure, redid data and evaluation, defined more real-world problems, improved data quality.
He especially emphasized anti-leaderboard thinking.
Yao said, leaderboards are valuable but prone to overfitting. Real user questions are often fuzzy, short, multi-turn, very different from precise benchmark tasks.
This explains why Tencent now emphasizes Co-Design.
Co-Design is the joint design of model and product.
The model is no longer hidden in the lab fine-tuning scores, but gets real feedback from products such as Yuanbao, WorkBuddy, CodeBuddy, then loops back to improve the model.
Tencent’s earnings call disclosed Hunyuan 3 Preview has been integrated into 131 Tencent products, current token calls at least 10x Hunyuan 2. Since April 28, it’s ranked first by token usage on OpenRouter, remaining in the lead after the free period ended on May 8.
This shows Hunyuan has at least moved from “not usable enough internally” to “can be massively used by products”.
But Pony Ma’s “standing but not able to sit” remark corresponds exactly here.
Hunyuan 3 Preview is progress, not the end.
Tencent management also says the next step is to develop models with larger parameters, using larger and higher-quality datasets and stronger reinforcement learning to improve contextual understanding, agent capabilities, and general intelligence.
This is the underlying variable for whether Tencent AI can continue accelerating.
Tencent’s Strength is Ecosystem, Challenge is Also Ecosystem
Tencent AI’s strongest card isn’t any single app.
But an entire ecosystem.
WeChat has users, WeCom has organizational ties, Tencent Docs and Meeting have office scenarios, mini-programs have merchants and services, WeChat Pay has transaction loop, ad systems have commercialization ability, Tencent Cloud has computing power and B-end clients.
This gives Tencent natural soil for agents.
But ecosystem is not automatic victory.
JP Morgan warns, for WeChat agent to enter higher-value stages, several issues must be solved: AI wallet integrated with main WeChat Pay wallet, agent-generated GMV must be discloseable, merchant migration as callable skills, agent user ad ARPU not deteriorating, regulation not restricting agent as business recommender.
None is easy.
Especially recommendation neutrality.
If WeChat agent helps users choose products, book services, find merchants, what ranks the recommendations?
If users think recommendations are “paid top placements”, trust is damaged.
If zero commercialization, Tencent can hardly get enough returns.
This makes WeChat AI more complex than WorkBuddy.
WorkBuddy mostly solves efficiency problems, payment logic is more direct.
WeChat AI has to handle transaction, ads, payments, merchants, fairness and regulation.
Its ceiling is higher, friction is greater.
“Great again” is too early, but “half a beat slower” is outdated
Now we can answer the original question: Has Tencent AI started to accelerate?
The answer: It has started.
But not in the way most outsiders expected.
Tencent didn’t choose to directly compete head-to-head with Doubao and Qianwen with a consumer chatbot.
It started advancing along two lines.
One is high-value productivity agents like WorkBuddy and CodeBuddy, prioritizing real tasks willing to pay.
The other is super-entry agents like WeChat “Xiaowei”, first validating ability in discovery, invocation, order construction and mini-program generation within WeChat ecosystem, then gradually exploring closed-loop transactions.
This path fits Tencent’s own conditions.
But it is also slower, heavier, and costlier.
It needs progress in models, product polishing, ecosystem coordination, cost reduction, and a workable business model.
So WorkBuddy is not proof of Tencent being “great again”.
It is more of a signal: Tencent AI has finally found a cut point connecting model, product, ecosystem and payment.
WeChat AI is not a magic wand instantly changing Tencent’s valuation.
It’s more like a progress bar: as user scale, payment integration, merchant migration, GMV attribution, and ad performance are gradually disclosed, the market will recalculate AI as part of Tencent’s valuation, not just an option.
Pony Ma’s remark is still applicable.
Tencent no longer feels the boat is leaking like a year ago.
But what really matters is not who shouts “great again”, but the next four metrics:
First, whether WorkBuddy can convert high usage into paid users and stable ARR.
Second, whether WeChat “Xiaowei” can move from assisting tasks to payment closed-loop.
Third, whether Tencent AI’s costs can be covered by ads, cloud, enterprise services, and subscription revenue.
Fourth, whether Hunyuan and WeChat WeLM model systems can synergize, not duplicate investments.
Risk Warning & DisclaimerThe market has risks; invest prudently. This article does not constitute personalized investment advice, nor does it take into account individual users' special investment goals, financial status or needs. Users should consider whether any opinions, views or conclusions in this article suit their particular circumstances. Invest accordingly at your own risk.