At a Tencent shareholder meeting in May, Ma Huateng made remarks that set a modest starting point for the company's AI ambitions. He noted that a year prior, they felt they had boarded the ship, only to find it was leaking. Now, they feel they are on board but cannot yet sit down comfortably, hoping the vessel would pick up speed.

At the time, public discourse was dominated by a single question: Was Tencent's AI development lagging? Its Yuanbao assistant gained traction later than Doubao, and its Hunyuan model was launched after Baidu's Ernie. Its AI products seemed to lack a strong presence.

However, within just a few months, the narrative around TENCENT's AI efforts appears to have completely reversed.

WorkBuddy is now frequently discussed, the WeChat AI assistant "Xiaowei" has begun internal testing, and analyst reports have started re-integrating Tencent's AI prospects into valuation models. Some investment communities even proclaimed over the weekend that "Tencent is becoming great again."

From a "leaking ship" to a sudden surge in popularity, has Tencent AI found its "new ticket"?

WorkBuddy's Breakthrough

A piece of data from late June surprised many.

Analysys data showed that WorkBuddy had a monthly visit count of 8.85 million, with 20 million monthly active users (MAU) and over 13 million daily active users (DAU). The DAU/MAU ratio was between 65% and 75%, a stickiness metric comparable to office tools like Slack.

The exact words from TENCENT's Q1 earnings report were: "Measured by daily active accounts, WorkBuddy has become the most popular efficiency AI agent service in China."

More noteworthy is the speed: user numbers grew by two orders of magnitude in three months, with 43 versions released in that period, including updates during the May holiday.

Reports suggest that internally, Tencent positions it as the "third phenomenon-level product" after QQ and WeChat, potentially the company's "new ticket in the AI era."

While this positioning has elements of internal promotion, the numbers themselves are compelling.

WorkBuddy's predecessor was CodeBuddy, an AI coding assistant for programmers. Development began in 2022. Starting from small-scale internal trials, by the end of 2025 it covered 90% of the company's engineers. Over 90% of code in most teams was AI-generated, reducing overall coding time by 40%.

In January this year, the team expanded it into a desktop agent for general users, with WorkBuddy version 0.01 internally incubated. It was officially launched domestically on March 9, with an overseas version released on May 28.

Key Factors for Success

Why has it succeeded? Two key points stand out.

First, it drastically lowered the barrier to entry. The problem with solutions like OpenClaw ("foreign lobster") was a high technical threshold—command-line deployment, self-funded tokens—which deterred non-technical users and even spawned a secondary market for installation services.

WorkBudby bypassed this entirely: a desktop client plus a WeChat mini-program with an extremely simple installation process. After WeChat authorization, users can send commands from their phone to automatically run tasks on their PC. This is a capability difficult for foreign counterparts to replicate.

Second, Tencent's ecosystem advantage has been activated in the Agent era. WorkBuddy has integrated Tencent Docs, Tencent Meeting, WeCom, Tencent Cloud Drive, and TAPD. It also connects to over 30 external tools like Feishu, Kingsoft, Qichacha, and SalesEasy via the MCP protocol. Its SkillHub library has expanded from 79,000 to 790,000 skills.

For example, telling it to "create a PPT with last week's sales data and send it to the meeting participants" can trigger it to pull data, generate the file, and dispatch it.

In the previous "chatbot" era, Tencent's ecosystem was fragmented. In the "get-work-done" Agent era, this suite of tools has become a moat. As President Martin Lau stated on the Q1 earnings call: "Users can control AI agents within communication and browsing interfaces like WeChat, QQ, and the Tencent browser."

WeChat AI: The Larger Card, Partially Revealed

Beyond WorkBuddy, an even larger card is being slowly revealed.

On June 20, 2026, the WeChat AI assistant "Xiaowei" began a limited gray-scale test.

What can Xiaowei do? Early feedback from test users indicates it can perform native WeChat operations via voice or text (sending messages, posting to Moments, making calls), invoke mini-programs for tasks like appointment booking and food ordering, conduct information searches and content generation, and even generate mini-program prototypes using natural language.

This marks the first foray into the agent arena for a super-app with 1.43 billion monthly active users.

However, it's important to assess its current state clearly.

In a July 6 research report, J.P. Morgan's team presented a three-tier value creation framework, clearly distinguishing between what is already feasible and what still requires waiting.

Tier 1 (a): Assisting the traffic funnel—Agents help users discover, compare, and build orders within mini-programs, but checkout still requires manual completion. This scenario is already supported in the current public test.

Tier 1 (b): Closed-loop transactions—Agents complete transactions end-to-end through integrated payment. This is not yet supported, as the AI wallet and the main WeChat Pay wallet are not yet connected.

Tier 2: Integrating a broader range of mini-program merchant supply, evolving WeChat from a traffic source to a semi-marketplace platform. The probability is rising, but merchant adoption rates remain to be validated.

Tier 3: WeChat becomes the dominant traffic entry point for online consumption, capturing a significant share of merchant lead-generation budgets. J.P. Morgan assigns only a 10% probability to this, citing unresolved issues like potential cannibalization of the ad business, neutrality of recommendation mechanisms, user trust, and regulation.

After probability weighting, J.P. Morgan forecasts that the WeChat AI Agent could generate incremental revenue of approximately 126 billion yuan by 2030, with incremental operating profit of about 88 billion yuan, representing roughly an 18% upside to their 2030 operating profit forecast.

Goldman Sachs, in a June 23 report, expressed a more cautious view on WeChat AI, highlighting three core concerns: First, Xiaowei uses WeChat's self-developed WeLM model, not the group's Hunyuan model, raising questions about potential resource duplication from running two large models. Second, if fully promoted, incremental inference costs could equate to approximately 5%-17% of Tencent's forecast adjusted operating profit for Q4 2026. Third, the short-term monetization path is unclear, with ad monetization requiring deeper AI penetration in local services, content discovery, and shopping scenarios.

Goldman's conclusion was: "The re-rating of Tencent's valuation multiples over the next few quarters will depend primarily on the progress of its AI narrative." They maintained a Buy rating with a 12-month target price of HK$700 (SOTP method), implying about 62% upside at the time.

The core disagreement between the two institutions lies in the timeline. J.P. Morgan believes the public test has turned WeChat AI from an "abstract option" into a "trackable milestone." Goldman Sachs sees short-term cost pressures and a lack of monetization symmetry, requiring more quarterly data for validation.

Competitive Reactions

Following the release of WorkBuddy's data, two of the most direct reactions came from competitors.

ByteDance: On June 9, Trae was upgraded from an AI programming tool to "Trae Work," with its slogan changed to "Real AI Enabler." It added Work/Code dual modes and began selling an enterprise version through Volcano Engine.

Alibaba: On July 2, it merged three products—QoderWork (office productivity), DingTalk "Wukong" (enterprise collaboration), and Alibaba Cloud MuleRun (Agent execution engine)—under the unified management of 1992-born Chen Yusen. On June 10, Alibaba's Partner Committee even posted internally, directly criticizing DingTalk's management culture as "not what Alibaba should be."

Leadership changes, mergers, and cultural critiques—these swift actions suggest they were genuinely stimulated by WorkBuddy's data.

The three companies are taking different paths: Tencent is pursuing "Agent + WeChat embedding," with a focus on government scenarios (already deployed for 2.3 million civil servants in Guangdong) and the WeCom foundation. ByteDance is moving from "programming tool → office → enterprise version," relying on its Doubao/Seed model family. Alibaba is leveraging its "DingTalk 700 million user base + Agent Marketplace."

Another notable detail: WorkBuddy is the first phenomenon-level product in Tencent's history to employ ground-level sales promotion—neither QQ nor WeChat did. This signal indicates Tencent's genuine commitment to penetrating the B2B market.

Hunyuan's Catch-Up: Rebuilding the Model-Product Cycle

The earlier criticism that Tencent's AI was slow cannot be viewed solely at the application layer.

The underlying model also underwent reconstruction.

In an official discussion, Yao Shunyu, speaking about Hunyuan 3 Preview, mentioned several key efforts: rebuilding infrastructure, reworking data and evaluation, defining more realistic problems, and improving data quality.

He specifically opposed a "leaderboard-chasing" mindset, noting that while benchmarks have value, they are prone to overfitting. Real-world user problems are often vague, short, and involve multi-turn questioning, differing greatly from precise benchmark questions.

This explains Tencent's current emphasis on Co-Design, where the model and product are designed together. The model is no longer locked in a lab chasing scores but receives real feedback from products like Yuanbao, WorkBuddy, and CodeBuddy, which in turn improves the model.

The earnings call disclosed that Hunyuan 3 Preview has been integrated into 131 Tencent products, with current token call volume at least 10 times higher than Hunyuan 2. Since April 28, it has ranked first by token usage on the OpenRouter platform and maintained its lead even after the free period ended on May 8.

This indicates that Hunyuan has at least progressed from being "not very useful internally" to being "capable of large-scale product calls."

However, Ma Huateng's comment about being "on board but cannot yet sit down comfortably" precisely corresponds to this stage. Hunyuan 3 Preview represents progress, not the final destination.

Management also stated that the next step involves developing larger-parameter models to enhance context understanding, agent capabilities, and general intelligence through larger-scale, higher-quality datasets and stronger reinforcement learning. This remains the foundational variable determining whether Tencent AI can continue to accelerate.

Tencent's Strength and Challenge: The Ecosystem

Tencent's strongest card in AI is not a single app, but its entire ecosystem.

WeChat has users, WeCom has organizational relationships, Tencent Docs and Meeting have office scenarios, Mini-Programs have merchants and services, WeChat Pay has the transaction loop, the advertising system has monetization capabilities, and Tencent Cloud has computing power and B2B clients.

This provides fertile ground for Tencent to develop Agents.

However, having an ecosystem does not guarantee automatic victory. J.P. Morgan cautioned that for the WeChat agent to enter higher-value stages, several issues must be resolved: integration of the AI wallet with the main WeChat Pay wallet, disclosure of GMV generated by agents, migration of merchants to provide callable skills, maintaining or improving Agent user advertising ARPU, and regulatory non-interference with agents acting as commercial recommendation intermediaries.

These are not simple tasks, especially the issue of recommendation neutrality. If the WeChat Agent helps users select products, book services, and find merchants in the future, what criteria will sort the recommendations? If users perceive recommendations as "paid placements," trust will erode. If commercialization is avoided entirely, it becomes difficult for Tencent to secure sufficient returns.

This is where WeChat AI is more complex than WorkBuddy. WorkBuddy primarily solves efficiency problems with a relatively straightforward payment logic. WeChat AI must handle transactions, advertising, payments, merchants, fairness, and regulation. Its ceiling is higher, but so is the friction.

Conclusion: "Great Again" is Premature, But "Slow" is Outdated

Now, we can answer the initial question: Has Tencent AI begun to accelerate?

The answer is yes, but not in the way most familiar to the outside world.

TENCENT did not choose to directly compete with Doubao and Tongyi Qianwen using a single consumer-facing chatbot. Instead, it is advancing along two lines. One involves high-value productivity agents like WorkBuddy and CodeBuddy, prioritizing real tasks that users are willing to pay for. The other involves super-entry agents like WeChat's "Xiaowei," first validating discovery, invocation, order building, and mini-program generation capabilities within the WeChat ecosystem, then gradually exploring closed-loop transactions.

This path is more aligned with Tencent's inherent strengths but is also slower, heavier, and more expensive. It requires model progress, product refinement, ecosystem coordination, cost reduction, and proven business models.

Therefore, WorkBuddy is not proof that Tencent is "great again." It is more like a signal that Tencent AI has finally found an entry point that connects models, products, ecosystems, and payment.

WeChat AI is also not a magic wand that will immediately change Tencent's valuation. It is more like a progress bar: as user scale, payment integration, merchant migration, GMV attribution, and advertising performance are gradually disclosed, the market will reincorporate AI from an "optional call option" back into Tencent's valuation.

Ma Huateng's words still apply. Tencent no longer feels like the ship is leaking as it did a year ago. However, what truly matters is not who proclaims "great again," but the following four indicators:

First, whether WorkBuddy can convert high usage into paying users and stable Annual Recurring Revenue (ARR).

Second, whether WeChat's "Xiaowei" can progress from assisting with tasks to enabling payment closed loops.

Third, whether Tencent's AI costs can be covered by advertising, cloud services, enterprise services, and subscription revenue.

Fourth, whether the Hunyuan and WeChat WeLM model systems can achieve synergy rather than representing duplicated investment.