Meituan's "Xiaomei" Connects to Tencent's Yuanbao: The Battle for Local Services AI Entry Points Begins Early

Two of China's internet giants have fired a major signal flare in the AI-powered local services arena. During its first-quarter 2026 earnings call, Meituan revealed that its AI assistant "Xiaomei" is about to complete integration with Tencent's AI assistant "Yuanbao." Users will soon be able to use natural language within Yuanbao to order food delivery and book restaurants, with Meituan handling fulfillment. This represents the first instance of "Agent-to-Agent" communication between two different companies' AI agents in China's AI industry, and signals that AI assistants are rapidly moving beyond the era of simple chatbots, racing to become super entry points connecting real transactions and lifestyle services.

Just one day after Meituan's announcement, multiple media outlets reported that ByteDance's Doubao had quietly integrated with Douyin's local services, enabling users to complete the full transaction process for group-buying deals, movie tickets, homestays, and more directly within Doubao. Almost simultaneously, the Financial Times disclosed that Tencent is testing an embedded AI agent prototype within WeChat, with compliance approval expected to begin as soon as this month. The密集 appearance of these three developments has pushed competition in China's AI application layer into a new phase defined by "conversation-as-transaction."

The Separation of "Entry Points" and "Capabilities"

Meituan CEO Wang Xing clearly articulated the company's AI strategy during the earnings call: "Meituan's true moat lies in its structural data advantage—covering all categories of local services, verified real-time information, authentic user reviews, and a powerful fulfillment infrastructure. By layering AI on top of these structural advantages, we will deliver the best AI-powered local services experience to users."

Wang Xing also specifically noted that beyond serving consumers and merchants, serving AI agents is becoming increasingly important. He described Meituan's posture in AI as "proactively attacking, rather than passively defending."

Behind this statement lies a fundamental shift in traffic distribution logic in the AI era. During the mobile internet era, traffic entry points were nearly synonymous with service capabilities—whoever had the users and the app controlled everything. But in the AI era, the entry-point layer and the application layer are accelerating their separation. AI assistants like Yuanbao, Doubao, and Kimi possess conversational entry points and massive user bases but lack actual fulfillment capabilities. Meituan has the deepest local services data, merchant network, and delivery infrastructure, but lacks an independent AI traffic entry point.

According to QuestMobile data, Yuanbao had 57.35 million monthly active users in March, maintaining high penetration in China's first-tier and new first-tier cities. By comparison, Doubao maintained a DAU of approximately 140 million after the Spring Festival peak, Alibaba's Qianwen held at around 30 million, while Yuanbao's DAU remained at only about 9 million. Tencent spent over 15 billion yuan (approximately $2.2 billion) on marketing for Yuanbao throughout 2025, but user retention has been less than ideal.

This scale gap means Yuanbao urgently needs real transaction scenarios to prove its value as an independent AI entry point. Meanwhile, Meituan needs a high-frequency AI conversational entry point to extend its service capabilities into users' daily conversation flows, rather than passively waiting for users to open the Meituan app.

A Complementary Deal Between "Body" and "Eyes"

The technical architecture of this cooperation is Agent-to-Agent: Yuanbao serves as the front-end conversational layer, responsible for understanding user intent; Xiaomei acts as the execution layer, responsible for calling upon Meituan's food delivery, dine-in, and delivery services. The two communicate task handoffs through standardized agent protocols, with users' raw data not directly flowing between the two systems.

The advantage of this architecture is clarity—Meituan does not need to open its core database to Tencent, and Tencent does not need access to Meituan's merchant review system. But the cost of clear boundaries is a ceiling on capabilities. If a user makes a complex request like "help me find a restaurant suitable for a weekend family dinner, around 150 yuan per person, with good reviews and convenient parking"—a query requiring deep understanding of Meituan's structured data—whether the standardized interfaces at the agent protocol level can support this remains unknown.

Cross-giant cooperation also faces two other core challenges. First, how to distribute commercial benefits. Every Meituan order completed through Yuanbao is essentially co-created by Yuanbao's traffic value and Meituan's service value. If the revenue-sharing ratio cannot be balanced, either party may lose the motivation for sustained investment. Second, whether the user experience can be truly seamless. The integration between Qianwen and Alibaba's local services is completed within the same corporate group, with account systems, payment chains, and verification logic naturally connected; the same applies to Doubao and Douyin's local services. But Xiaomei and Yuanbao are building a bridge between two separate companies—when a user issues a command in Yuanbao, jumps to Meituan to complete payment, and encounters an issue, which customer service do they contact? These connection details ultimately determine whether the experience is merely "functional" or genuinely "excellent."

Setting a Template for WeChat's AI Agent

The timing of this cooperation is intriguing. Just one day after Meituan's earnings release, the Financial Times disclosed that Tencent is testing an embedded AI agent prototype within WeChat, which has been designated as the company's highest strategic priority. Tencent's stock price surged more than 10% that day, marking its largest single-day gain since November 2022.

According to the Financial Times, the interaction entry point for this agent is remarkably direct—users simply swipe right on WeChat's main interface to bring up a conversation window. After inputting commands, the agent can automatically call upon WeChat's vast ecosystem of mini-programs to complete various tasks. Tencent plans to initiate compliance approval this month, followed by a small-scale external user灰度 test before a phased full rollout.

However, the core value of WeChat's AI agent lies in connecting millions of mini-programs, but major platforms within the mini-program ecosystem—such as Meituan, JD.com, Pinduoduo, and Ctrip—have their own data sovereignty and commercial interests and will not unconditionally hand over their service capabilities to WeChat's agent for调度. Tencent cannot avoid a fundamental question: why would the big players be willing to connect?

The cooperation between Yuanbao and Xiaomei is precisely an early test of this question. Are major internet platforms willing to connect to Tencent's AI ecosystem via "Agent-to-Agent" protocols? Meituan's answer is "yes"—the interests of both parties are complementary. Meituan gains an entry point, Yuanbao gains a scenario, and neither side is forced to cede core rights. This logic of "complementary cooperation" is exactly the key reference for whether the WeChat AI agent ecosystem can recruit more ecosystem partners going forward.

The future WeChat AI agent ecosystem will likely differentiate into three tiers: the first tier is "Agent-to-Agent," where platforms like Meituan that possess deep vertical data but lack independent AI entry points connect to the WeChat agent via agent protocols, each retaining data sovereignty and brand independence. The second tier is "Agent-to-Mini-Program," where a large number of small and medium-sized merchants without the capability to build their own agents respond to WeChat agent调度 by upgrading their mini-program interfaces. The third tier is Tencent's self-operated businesses—products like Tencent Maps, Tencent Health, and Tencent Video will be directly and deeply integrated by the WeChat AI agent.

ByteDance and Alibaba's "Walled Gardens"

The cooperation between Meituan and Yuanbao is, to some extent, a choice forced by competitive pressure.

According to reports from Sanyi Life, after previously integrating with Douyin e-commerce to support in-app shopping and order completion, Doubao recently further connected with Douyin's local services—users can now use natural language within Doubao to express their needs, receive store and group-buying package recommendations, and complete the full process including payment, order viewing, and verification code retrieval. Categories currently live include food and dining, movie tickets, and homestays. This means Doubao has already built a complete "conversation-as-transaction" closed loop, backed by the merchant resources and fulfillment system that Douyin's local services have accumulated over years of deep cultivation.

Alibaba's Qianwen is similarly accelerating the construction of an AI-era "walled garden," deeply integrating with Alibaba's local services. Both ByteDance and Alibaba are building closed commercial service systems and will not easily open调用 interfaces to external底层 services. The only AI assistant truly willing to complement rather than compete with Meituan is essentially Tencent's Yuanbao.

Meituan's loss-reduction performance in Q1 also indirectly confirms that its moat remains solid. During the quarter, Meituan's core local commerce operating loss narrowed to 2 billion yuan (approximately $295.6 million), a significant improvement from 10 billion yuan (approximately $1.5 billion) in the previous quarter. Losses from new businesses, including Xiaoxiang Supermarket and Keeta, also narrowed from 4.6 billion yuan (approximately $679.9 million) to 2.1 billion yuan (approximately $310.4 million). Boosted by the loss-reduction news, Meituan's stock price surged 9.27% after the earnings release, with a market capitalization of approximately HK$527.9 billion (approximately $67.4 billion).

Wang Xing noted during the earnings call that Meituan's significant loss reduction in Q1 was primarily due to its advantage in the mid-to-high average ticket price dine-in market remaining largely unshaken. "This is real demand that has survived the洗礼 of food delivery subsidy wars, and it is Meituan's most solid moat."

Structural Reset in the AI Application Layer

The cooperation between Meituan and Yuanbao is taking place against the backdrop of a structural reset in China's AI application layer.

According to Forbes Japan, ByteDance recently cut approximately 30% of its AI application projects, explicitly abandoning its strategy of "launching a thousand products in all directions." ByteDance's AI inference costs in 2025 exceeded 8 billion yuan (approximately $1.2 billion), approximately 2.3 times its incremental AI product revenue. A ByteDance executive stated bluntly: "At this rate of capital consumption, it's impossible to produce another large-scale application like Doubao."

This is the opposite of the mobile internet era's logic. In the mobile era, the larger the scale, the closer marginal costs approached zero, and the stronger the profitability. But in the AI era, every user query generates real computation and storage costs, with costs concentrated primarily on the service provider side. Growth does not dilute costs—it deepens losses.

Forbes Japan noted that China's AI application market is polarizing: on one pole are ultra-large-scale players like ByteDance, OpenAI, and Google, racing to build general-purpose platforms for massive user bases—capital-intensive, high-risk, with diminishing marginal returns. On the other pole are quieter but growing "precision AI" systems, built for specific workflows and loyal customer bases, pursuing depth over breadth.

Against this backdrop, the cooperation between Meituan and Yuanbao offers a unique path: not pursuing becoming an independent AI entry point, nor a closed "walled garden," but rather positioning as a "底层 capability provider," opening local services to external AI assistants through Agent-to-Agent protocols. Just as WeChat Pay became the infrastructure for e-commerce, Meituan is attempting to become the "water, electricity, and gas" of local services in the AI era—any AI assistant wanting to serve users' daily needs for food, clothing, housing, and transportation may ultimately find it difficult to bypass Meituan.

Meituan's investment footprint in the AI space is also considerable. From AI large model companies like Zhipu and Moonshot AI, to embodied intelligence firms like Unitree Robotics and Galaxy Universal, to computing power chip players like Moore Threads and Muxi, Meituan stands behind a significant portion of China's AI unicorns. While these investments are more about future ecosystem positioning, they provide Meituan with the confidence to play an "infrastructure" role in the AI era.

Wang Xing's concluding remarks during the earnings call were意味深长: "In the AI field, Meituan always chooses to proactively attack, not passively defend." This handshake between "Xiaomei" and "Yuanbao" may not be large in scale, but its positioning is critical—it occurs just before the WeChat AI agent officially takes off, and within the highest-frequency local services transaction scenarios of China's mobile internet. If this step works, the story that follows will have something to tell.