After Meta, Big Tech Awaits a "Muse Moment"

In September of this year, Meta launched the personal agent Muse, aimed at automating a range of tasks for consumers, including shopping, buying movie tickets, and scheduling appointments.

At Meta's latest third-quarter earnings call, CEO Zuckerberg said that Muse will be commercialized earlier than the company's previous new products. In addition to the existing paid subscriptions, he also wants to collect fees from transactions facilitated by Muse, which are expected to be paid by merchants. Zuckerberg even predicts that Muse will ultimately serve "billions of users," and the potential revenue from these users may exceed Meta's average user level.

Previously, it was difficult for the market to answer one question: when a user chats with AI for an hour, how much is that time actually worth? Muse provides the answer: AI helps users compare prices, make appointments, make purchases, and place service orders, and value is realized once the transaction is completed. This logic happens to provide a frame of reference for domestic AI giants that are currently in a period of uncertainty.

Since the beginning of this year, shutdowns, mergers, downgrades, and consolidation of entry points have become the main theme of domestic giants' AI products. It is not that AI demand has disappeared, but that the cost-no-object, scattergun Agent 1.0 experiment has officially come to an end.

Less than two months after launch, Meta is already talking at length about Muse's monetization strategy and prospects, and it even plans to commercialize it earlier than the company's previous new products.

Muse indeed has grounds for confidence. According to management disclosures, Muse's most common use case is shopping; more than a quarter of users complete purchases through it, with an average order value of more than $200. Transaction volume in the first week after launch exceeded internal expectations, requiring cooperation with Stripe to raise credit limits. About 25% of users who have set up Muse use it daily, with retention more than twice that of Threads in its early stage. The company estimates that at least 500 million small and micro businesses acquire customers through Meta-family apps, and Muse can simultaneously serve merchant customer acquisition and user transactions, with the commercialization path initially taking shape.

According to previous statements, Muse will insist on being ad-free, ordinary users can use it for free, but advanced features and enterprise services will be charged. Meta will profit through multiple channels: transaction fees paid by merchants, subscription fees paid by heavy users and customers who want to use the most advanced models, and paid enterprise tools for businesses that rely on Muse for work.

At present, Meta has not yet announced the commission rate, the scope of applicable merchants, or the official launch time. The final revenue scale will depend on Muse's task execution capabilities, merchant onboarding, and whether users are willing to hand shopping decisions to an agent. In the future, Meta will not only develop more types of Muse, but also connect Muse to more devices, such as smart glasses and the dedicated device Charm, using multiple touchpoints and entry points to encourage more people to be willing to pay continuously.

Zuckerberg said at the earnings call: "We did not directly charge everyone upfront, and one reason is that we want to build the Muse network. We believe that when a massive number of users and businesses are using Muse, its value at scale will be higher." He also added: "If compute efficiency continues to improve rapidly in the coming years—and we think that is highly likely—then Muse will generate substantial profits for many years to come."

Judging from this earnings report, most of Meta's revenue still comes from advertising. But Meta Chief Financial Officer Susan Li expects that the company's AI business will begin contributing incremental revenue starting in 2027 and, over time, become "an important source of revenue diversification."

Compared with the past two years, the agent platforms of domestic giants have mostly remained stuck outside the inherent model of "create—distribute—no one pays." Muse provides a consumer-facing revenue stream that can genuinely reflect the output of large investments. Even if the scale of commercialization is small now, Muse's significance lies in establishing a measurable causal chain for AI investment: AI investment → changes in user behavior → transaction commissions → revenue growth. For example, a series of indicators such as millions of subscribers, billions of dollars in gross merchandise value, app downloads, and user activity can all be used to measure the growth of the consumer AI business.

This is also why Meta's stock price was boosted after the earnings report—largely because investors saw the prototype of this chain. The market no longer only cares about how much it spends building data centers, but has begun paying for the incremental revenue brought by AI applications.

Domestic giants' "cutting loose": shutting down platforms and betting on new entry points

In the same week that Meta announced Muse's commercialization, Tencent's Yuanqi team issued an announcement that the platform will stop all public services on November 9. Previously, Tencent QClaw had announced that it would cease operations at the end of the year. The creation entry point of Baidu's Wenxin Agent Platform was also closed in August. ByteDance's moves came earlier and more thoroughly: in July, the Feishu product team merged with the Doubao team, with the head of Feishu reporting to the head of Doubao, while the commercialization team was merged into the Volcano Engine system. The AI short drama product Xiaoyunque was also recently reported to be merged into Jimeng. Alibaba Cloud also recently officially announced the delisting of its virtual digital human.

The logic behind this round of AI product shutdowns, transfers, and mergers is not hard to understand, mainly due to three reasons: first, compliance pressure. In July 2026, the Interim Measures for the Administration of Anthropomorphic AI Interactive Services officially took effect, bringing open UGC agents under strict regulation. Second, cost pressure. Standalone products mean duplicated investment in computing power, R&D, and operational resources; many products only obtain user retention data but no revenue, making it hard to sustain before commercialization is proven. Third, strategic focus. In the past, major companies collectively launched agent platforms, zero-code tools, and standalone Agent applications, with a large number of products addressing similar needs. As foundation model capabilities rapidly converge, thin wrapper-layer products are easily replaced by native capabilities of the main app. Now the major companies are consolidating scattered capabilities into core entry points to avoid "multi-front warfare."

While the major companies are doing "subtraction," they are also doing "addition"—betting on personal agents. According to media reports, in laying out personal agents, ByteDance's "Spell" project is led by the Doubao mobile assistant team, has been in internal beta since April, and is planned to launch after deep integration with Doubao; Tencent launched LightVela, and the WeChat AI assistant "Xiaowei" has entered internal beta; Alibaba's Qwen App has officially launched "Qwen Go!", attempting to handle matters such as flight claims, shopping price comparisons, and phone appointments; Ant's "Lingguang" is also in small-scale internal beta for personal agents.

Previously, most domestic personal agents remained stuck in the vague positioning of an "all-purpose assistant," wanting to serve both C-end users and the B-end ecosystem, and the result was often falling short on both ends.

Muse's success has also given this batch of new domestic personal agents a clearer reference and established a new product development logic: no longer building generalized open agent plazas, but binding to clear user groups and fixed high-frequency scenarios, embedding commercialization paths from the outset, with the goal of seizing the next-generation AI entry point at the phone, system, and hardware layers.

English text is being prepared.

A CICC research report also points out that the C-end ChatBot model has relatively limited value, while Agentic AI offers broader space; in the long run, the C-end will most likely adopt a "free + performance advertising" model, and performance advertising in the Agent era will be more transaction-oriented (CPS model).

1. In MaaS/API services, leading vendors charge by call volume or tokens, which is currently the most stable source of revenue;

2. For C-end subscriptions, Doubao, Wenxin, Qwen, Kimi, and others have successively launched paid memberships, exchanging higher context, faster reasoning, and exclusive Agent quotas for stable cash flow;

3. For results-based payment (RaaS), customers pay for actual delivered quantifiable business outcomes, such as e-commerce marketing taking a 5%-15% share of GMV, and financial risk control taking a 10%-20% share of loss reduction;

4. For ecosystem revenue sharing, platforms take a 15%-30% share from developer or transaction revenue;

5. For overseas commercialization, domestic models are accelerating overseas expansion with cost advantages, providing services to overseas B-end customers, developers, and C-end users, using overseas payment habits to offset the low C-end payment conversion rate domestically, becoming an incremental track—for example, DeepSeek and Zhipu have successively connected to the settlement systems of global leading platforms such as AWS.

In its second-quarter report this year, when Tencent mentioned Xiaowei, it also mentioned how to build a closed loop between agents and agent transactions—that in the future users would only need to give a dedicated agent one complex instruction, and the agent could automatically complete the entire transaction process. But Tencent's wording on this was also cautious, emphasizing "user experience" and "privacy" rather than the pace of monetization.

If comparing domestic and overseas agent business paths, overseas Muse may more easily prove out the transaction closed loop of standalone Agents first, while domestically companies will rely on existing e-commerce, local life, and enterprise service ecosystems, pilot commission models in small steps, use subscriptions as a backstop, and slowly turn AI from a "chat tool" into a trusted digital agent.

Regardless of the process, the AI commercialization companies that ultimately succeed need to satisfy three things at the same time: controllable computing power costs, stable task delivery capability, and a value chain that can be rights-confirmed and attributed. Relying only on membership subscriptions or only on API calls will have a ceiling in the long run; players that can run a combination of "subscription backstop + result sharing + B-end productivity solutions" will gain pricing power over the next-generation AI entry point.

Conclusion

Before last year, the narrative of the global AI industry was "invest—catch up—invest again." This year, the narrative has become "who can first find a billable closed loop." Meta gave an answer with Muse: make AI an intermediary for transactions, not a porter of content. Merchants' willingness to pay for completed transactions is far stronger than their willingness to pay for exposure. This logic held in the advertising era and will only be stronger in the agent era.

Domestic major companies are undergoing a shift from "laying out tracks" to "calculating ROI." Shutting down inefficient products, consolidating core capabilities, and betting on personal agents is essentially directing limited resources toward areas with clearer commercialization paths. In the next stage, whoever can first run through the combined model of "subscription + transaction commission + ecosystem revenue sharing" will gain an early advantage in the battle for the AI entry point. And computing power efficiency, compliance capability, and ecosystem closed loops will become the three key variables determining victory or defeat.

This article comes from the WeChat public account "Business New Research Society," author: Shang Jing, published by 36Kr with authorization.