Meta's Chief AI Officer Alexandr Wang: Muse Integrates with Walmart and Other Retailers
Alexandr Wang announced at the Connect conference that the personal assistant Muse will integrate with Walmart, Best Buy, Gap, Sephora, Wayfair, as well as Dick’s, Ulta, and Fanatics, with payment options including Shop Pay and PayPal. The work side will integrate with Box, GitHub, Granola, and Notion. Zuckerberg stated that basic usage will remain free for a large number of tokens, with paid tiers at $20 and $100 per month. Amazon continues to block the assistant from browsing its site.
Muse launched on September 8, targeting users over 18 in the U.S., available through a standalone app, web, and WhatsApp. Wang cited users saving $85 a month on cable bills, negotiating insurance, and tracking refunds. The product will feature a dedicated email, an operable Mac application, and will enter AI glasses in the coming months, allowing products on shelves to be identified without verbal description. The open connector received over 1,500 developer applications within a week. Morgan Stanley estimates the assistant shopping market could reach up to $30 trillion, while JPMorgan raised its target price from $820 to $920.
Wang became Chief AI Officer after Meta acquired approximately 49% of Scale AI for $14.3 billion in 2025, overseeing the Super Intelligence Lab. In April 2026, the lab will deliver its first closed model, Muse Spark, with officials stating it took nine months to overhaul the technology stack. After Yang Likun's departure in November 2025, he told the Financial Times that Wang had "no experience" and did not understand how to conduct research. Meta also formed an engineering team not under his management, and the company used a group photo to dispel rumors of his "fall from grace." Scale started with data labeling, with clients including Waymo, the military, and large model labs; Wang entered YC with Lucy Guo at 19, without completing his studies at MIT.
Meta is shifting from taking a cut of ad clicks to taking a cut of transactions made through the assistant. After placing an order through the assistant, the platform takes a cut, with retailers exchanging access for entry points, and users exchanging dialogue for fulfillment. The connector hands inventory and payments to a sandbox virtual machine, and Amazon's absence indicates that the platform can still sever access to shelves. The free tier swaps for daily active users, while the $20 and $100 tiers cater to heavy automation.
In market mechanics, what is sold is an assistant that can negotiate and place orders, while what is bought are retail APIs and payment channels. Demand comes from consumers who do not want to compare prices themselves, and supply comes from brands that want to appear in the assistant's recommendations. The beneficiaries are Meta, which takes a cut from transactions, and retailers set as default shelves, while the pressured parties are Amazon's blocked search ads and traditional comparison sites. Funds flow from product prices through Shop Pay and PayPal to merchants, and then back to Meta according to revenue sharing.
Source: Public Information
ABAB AI Insight
Zuckerberg did not buy a stake in a labeling company for $14.3 billion; he acquired an operator capable of rebuilding the lab according to delivery cycles. After the controversies surrounding Llama 4's reputation and ranking, the open-source route has given way to the closed Muse series. Wang translated Scale's "people feeding data to models" into "models executing for people," turning label workers into Walmart inventory on the other end of the connector. Yang Likun's departure, with over $1 billion in financing for world models, represents a fork in research routes, not the end of personnel gossip.
The capital path connects the advertising ledger to the assistant ledger. Instagram and Facebook direct traffic to Muse, while glasses and the Muse Charm pendant shift requests from screens to sight. Taking a cut from transactions is closer to payment companies and relies more on whether retailers are willing to open interfaces. Amazon's blockade proves that shelf rights can still veto everything; Walmart's integration shows that someone is willing to exchange traffic for toll fees.
Similar structures can be seen in Apple's App Store cuts, Google's search ads, and PayPal's checkout fees. Meta is in the stage of translating daily active users into assistant GMV: Spark proves the model can be presented, Connect shows that merchants are willing to connect, and glasses will determine whether requests leave the phone.
Structural judgment indicates a transfer of pricing power. The shopping entry point shifts from search boxes and shelf apps to assistant dialogues. The mechanism is: whoever controls the executing assistant and default connector will charge brands a new listing fee; those who started with labeling will change from "telling the model that is a pedestrian" to "placing orders for the goods needed by pedestrians."
ABAB News · Cognitive Laws
- Those who started with labeling will eventually include execution rights in contracts.
- Assistants can place orders, and shelf rights are harder than model scoring.
- Naivety has no roadmap, so it dares to overhaul the technology stack.