MENLO PARK — Meta’s newest artificial-intelligence agent is moving beyond answering questions and into something potentially much more disruptive: spending money on behalf of users.
Meta’s Muse personal AI agent can search online stores, compare products and prices, present recommendations and — after receiving a user’s approval — complete purchases on supported websites. Meta is positioning the technology as a major step toward what Silicon Valley increasingly calls “agentic commerce,” where AI systems do much of the shopping work traditionally performed by consumers themselves.
The system can go considerably further than simply finding a cheaper pair of shoes.
Meta says Muse can also monitor prices, track purchases, negotiate for used goods on Facebook Marketplace, initiate returns, request refunds and communicate with merchant support systems.
A Meta spokesperson told CNBC that shopping has rapidly emerged as one of Muse’s largest usage drivers since the personal AI agent launched in September.
That makes the stakes much bigger than another chatbot feature.
If consumers eventually trust AI agents to decide where to shop, which product to choose and which retailer gets the transaction, the technology could reshape one of the most valuable parts of the internet economy:
the moment when a consumer decides who gets the sale.
Muse Doesn’t Just Recommend a Product — It Can Buy It
The shopping workflow is designed to be conversational.
A user can ask Muse for something like a pair of shoes, a laptop or a household product while specifying:
a preferred brand,
a maximum price,
a color,
size,
delivery requirements,
or other preferences.
Muse then uses its browser to search retail websites and identify products matching those instructions.
The agent displays potential purchases inside the Muse interface.
Depending on the retailer, the user can either be sent directly to the merchant’s site or allow Muse to continue toward checkout.
The major difference from ordinary online search is what happens next.
Instead of the shopper manually navigating through multiple sites, comparing products, entering information and clicking through checkout screens, the agent can perform much of that process itself.
The customer retains final approval before the transaction is completed, Meta says.
Meta Says Your Real Card Number Can Stay Hidden
Payments are one of the most sensitive parts of autonomous shopping.
Meta says Muse can generate a one-time virtual card number during checkout, preventing the merchant — and the agent workflow — from exposing the user’s actual card number directly.
That feature is designed to address one of the biggest obstacles facing AI shopping agents:
trust.
Consumers may be comfortable asking an AI where to buy headphones.
Allowing the same software to actually spend money creates a much higher psychological and security barrier.
Meta is therefore emphasizing that customers remain in control.
Muse can prepare and execute tasks, but sensitive actions such as purchases require approval.
Muse Can Even Handle Returns
The technology does not stop once the package arrives.
Meta says Muse can interact with merchant customer-service systems to:
begin returns,
request refunds,
file disputes,
and follow up until the issue is resolved.
Again, the company says the user maintains approval authority.
This could prove especially attractive because returns are one of online shopping’s most frustrating tasks.
Customers often have to locate order numbers, find return policies, talk with chatbots, print labels or repeatedly follow up with support staff.
An AI agent capable of handling those repetitive steps could save substantial time.
But it also means AI agents may increasingly become the primary interface between merchants and customers.
That could have major consequences for retailers.
Facebook Marketplace Could Become Much More Automated
Muse also integrates directly with Facebook Marketplace.
Meta says the agent can monitor listings, alert users when matching items appear and even make offers or negotiate with sellers on a shopper’s behalf.
It can work in the other direction as well.
Muse can help people list their own items for sale.
That is particularly significant because Marketplace is one of Meta’s largest existing commerce products.
Historically, transactions there required substantial manual interaction:
searching listings,
messaging sellers,
negotiating,
arranging pickup,
and handling follow-ups.
An agent that automates those steps could make Marketplace significantly more efficient — while giving Meta another powerful reason to keep users inside its ecosystem.
Five Million Downloads in Just Weeks
Muse has attracted unusually rapid early adoption.
Reuters Breakingviews reported that the app had surpassed 5 million downloads by the end of September, only weeks after its September 8 launch.
Meta describes Muse as a personal AI agent, rather than simply an AI chatbot.
The distinction is important.
Chatbots answer questions.
Agents perform actions.
Meta says Muse runs inside a dedicated virtual environment called Muse Secure VM, giving the agent its own browser and allowing it to work across online services on behalf of the user.
That means Muse can do things such as:
book appointments,
fill out forms,
manage customer-service requests,
send messages,
book travel,
and shop.
The broader vision is for users to tell Muse what outcome they want rather than manually completing every individual step.
Zuckerberg Wants Muse Everywhere
Meta is not treating Muse as a standalone smartphone app.
At Meta Connect 2026, CEO Mark Zuckerberg announced plans to bring the agent to Meta’s AI glasses, allowing users eventually to ask Muse to take action based on what they are physically looking at.
For example, someone could look at a product on a store shelf and ask Muse to determine whether it is cheaper online.
Or the user could look at a school-supplies list and ask the agent to find and purchase the required items.
Meta says Muse will eventually work through more than 100 styles of AI glasses across Ray-Ban, Oakley and Meta-branded hardware.
That creates a very different commerce model.
Instead of pulling out a phone, opening a shopping application and searching manually, a person could simply look at something and tell their AI agent:
find this cheaper.
Shopify Could Give Muse Access to Millions of Products
Meta launched Muse with dozens of service integrations and access to Shopify’s product catalog, dramatically expanding what the agent can search and buy.
That partnership matters because Shopify powers millions of merchants around the world.
If Muse becomes deeply integrated with Shopify stores, Meta could instantly gain access to a massive ecosystem of independent retailers without building its own Amazon-style inventory network.
That could allow Meta to compete in online commerce in a very different way.
Amazon owns the marketplace.
Meta may try to own the shopping agent.
Instead of convincing sellers to list products inside one Meta-controlled store, Meta can send its AI agent across the open web to locate them.
Some Major Retailers Are Welcoming AI Agents
The retail industry is deeply divided over the technology.
The Wall Street Journal reported that companies including Walmart, Gap and Best Buy have embraced or experimented with Muse-style shopping agents, while Shopify intends to make its merchant ecosystem compatible with agentic commerce.
Supporters see the technology as another customer-acquisition channel.
If an AI agent recommends a retailer’s product, the retailer gains a sale it might otherwise have missed.
AI could also reduce cart abandonment by making checkout easier.
For merchants willing to adapt, agents could become the next major traffic source after:
Google search,
social media,
online marketplaces,
and influencers.
But not every retailer is convinced.
Amazon and eBay Have Reportedly Shut the Door
Some of the world’s largest e-commerce businesses are much more cautious.
The Wall Street Journal reported that Amazon and eBay have blocked or restricted shopping agents, citing concerns including security, customer data and control over the buying experience.
Other retailers, including companies linked to Tapestry and Kohl’s, have also imposed restrictions on automated purchasing.
The concern is understandable.
If customers increasingly shop through Muse rather than visiting Amazon, eBay or another retailer directly, the retailer risks losing control over:
advertising,
recommendations,
customer data,
brand presentation,
and loyalty.
A customer might technically buy something from Amazon.
But if Muse selected the product, compared it with rivals and initiated the purchase, who really owns the customer relationship?
That may become one of the defining battles of online commerce.
AI Agents Could Destroy the Value of ‘Consumer Inertia’
Reuters Breakingviews argues Muse threatens companies whose profits depend partly on customers being too busy, distracted or reluctant to comparison-shop.
Consider subscriptions.
Millions of consumers continue paying for services because canceling requires effort.
An AI agent can identify the subscription and cancel it.
Consider insurance.
Consumers often renew policies without comparing dozens of alternatives.
An AI agent can compare them automatically.
The same logic applies to:
travel,
fitness memberships,
bank accounts,
car purchases,
media subscriptions,
and retail products.
Reuters noted that shares of companies across travel, financial services, fitness, auto retail and entertainment came under pressure as investors considered whether powerful AI agents could strip away the margins created by inconvenience and customer inertia.
That is the potentially revolutionary part of agentic commerce.
AI agents do not get tired of comparison-shopping.
Humans do.
Travel Companies Have Already Felt the Fear
Companies such as Booking Holdings and Expedia have faced investor concerns that AI agents could weaken their position as intermediaries.
Traditionally, consumers open a travel site to compare flights and hotels.
An AI agent could potentially search across dozens of platforms without requiring the consumer to visit any of them directly.
The same principle applies to shopping.
If Muse becomes the interface consumers trust, individual retail websites risk becoming invisible suppliers behind the AI agent.
The shopper may remember Muse.
Not necessarily the merchant.
Meta’s Advertising Empire Creates an Even Bigger Question
Meta’s current business generates the overwhelming majority of its revenue from advertising across Facebook and Instagram.
That creates an obvious question:
How will product recommendations inside Muse remain neutral if Meta can make money from merchants?
For now, Meta’s public shopping materials emphasize convenience, user control and transaction capability rather than sponsored placement.
But Business Insider reported that Zuckerberg has discussed transaction fees as one possible future business model for Muse. Analysts have also questioned whether advertising could eventually become an important monetization channel.
That raises potentially difficult conflicts.
If two products are similar, does Muse recommend the genuinely best choice?
Or could the retailer paying Meta more receive preferential placement?
The answer will matter enormously as shopping agents scale.
Meta May Take a Cut of Transactions
One obvious business model is transaction fees.
Instead of charging customers directly, Meta could collect a small amount whenever Muse completes a purchase.
At massive scale, even tiny fees could generate substantial revenue.
But Business Insider cited analysts questioning whether consumers shop frequently enough for transaction commissions alone to justify Meta’s enormous AI investment.
The average consumer may make relatively few online purchases every month that an autonomous agent can realistically handle.
That means Meta could eventually need several revenue streams.
Possibilities include:
merchant fees,
transaction commissions,
subscriptions,
advertising,
or premium agent services.
No final long-term business model has been established publicly.
That Matters Because AI Is Becoming Extremely Expensive
Silicon Valley is spending extraordinary amounts of money on AI infrastructure.
Reuters reported that the global technology industry is now confronting a fundamental economic question: whether AI can create enough new revenue to justify the trillions being invested in data centers, chips and energy infrastructure.
Meta faces exactly that challenge.
Building models and operating agents that browse the web, perform tasks and run continuously requires enormous computing resources.
Personal AI agents therefore cannot remain expensive experiments forever.
Eventually, companies will need to monetize them.
Shopping may be one of the first opportunities capable of producing direct revenue.
Muse Is Also Moving Into Business Software
Meta is simultaneously expanding Muse beyond consumer tasks.
In September, the company launched Meta Enterprise Platform, bringing Muse and related AI agents into business environments.
Meta also introduced Muse for Small Business, which connects with platforms including:
Shopify,
QuickBooks,
Slack,
Stripe,
Canva,
Dropbox,
Notion,
Zoom,
and Facebook and Instagram business accounts.
Businesses can ask the agent to perform tasks such as finding customers and helping operate workflows.
Meta says nothing is published, sent or spent without the user’s approval.
The strategy shows how broad Zuckerberg’s vision has become.
Muse is simultaneously trying to become:
a personal assistant,
a shopping agent,
a business employee,
and eventually a hardware interface.
The Privacy Question Is Much Bigger Than Shopping
The more useful an AI agent becomes, the more personal information it needs.
Wired reported that researchers examining Muse found the agent creates evolving profiles of people in users’ lives, potentially including relationships, shared history and other personal context.
Meta says this information allows Muse to become more useful and personalized, while users retain controls over their data and agent actions.
But privacy experts have raised concerns about how deeply personal agents could understand users compared with traditional applications.
Shopping makes that issue particularly sensitive.
A shopping agent could potentially learn:
what users buy,
how much they spend,
what brands they prefer,
where they travel,
what gifts they purchase,
and what financial tradeoffs they make.
That kind of information could be extraordinarily valuable for personalization.
It could also become some of the most commercially sensitive consumer data in the technology industry.
Consumer Trust Is Still the Biggest Barrier
Technology companies are racing ahead faster than consumer comfort.
The Wall Street Journal reported that surveys still show relatively limited willingness among consumers to allow AI systems to make autonomous purchases.
That gap matters.
Finding a product with AI feels low risk.
Buying it automatically feels different.
Negotiating with strangers feels different again.
Giving an agent access to financial accounts may represent an even larger leap.
Meta can build the technology.
The harder task may be convincing users to trust it enough to hand over meaningful authority.
Muse Could Become a Major Threat to Google Search
Shopping is also one of Google’s most valuable search categories.
Consumers routinely search phrases such as:
“best laptop under $1,000”
“cheap flights to Tokyo”
“best running shoes”
or
“iPhone deals.”
Those searches generate valuable commercial advertising revenue.
But an autonomous agent changes the workflow.
Instead of searching Google, opening five websites and comparing prices, a consumer could simply ask Muse:
Find me the best laptop under $1,000 and buy it when the price drops below $850.
The user may never see the Google search results.
That is why AI agents could threaten not only traditional retailers but also the advertising intermediaries that helped consumers discover those retailers.
Meta Already Owns Consumer Attention
Meta enters this fight with an enormous advantage.
Billions of people already use:
Facebook,
Instagram,
Messenger,
and WhatsApp.
Muse can operate directly inside WhatsApp in addition to its standalone application.
That means Meta does not necessarily have to convince users to adopt an entirely new communication habit.
It can place the agent inside apps they already use every day.
That distribution could become just as important as the quality of the underlying AI model.
The Next E-Commerce War May Not Be About Stores
For decades, online retail competition centered on marketplaces and websites.
Amazon wanted shoppers to begin at Amazon.
Google wanted shoppers to begin with search.
Instagram wanted users to discover products through creators and ads.
Shopify wanted merchants to control their own storefronts.
AI agents could change the starting point completely.
Instead of choosing the store first, a consumer chooses the agent.
Then the agent chooses the store.
That small change transfers enormous power.
The company controlling the AI agent could influence:
which merchants get discovered,
which prices consumers see,
which brands receive recommendations,
and ultimately where trillions of dollars in online spending flow.
The Agent Could Become More Valuable Than the Marketplace
That may be Meta’s largest opportunity.
It does not necessarily need to build another Amazon.
It may only need to control the software that decides whether the customer buys from Amazon, Walmart, Shopify, eBay or somewhere else.
If Muse becomes trusted enough, Meta could sit above the entire retail ecosystem rather than compete inside it.
That helps explain why merchants are simultaneously excited and nervous.
Muse could send them customers.
But it could also place a powerful intermediary between the retailer and those customers.
And as online commerce shifts from humans clicking through websites to autonomous agents negotiating purchases with other software, the most valuable position may no longer belong to the company selling the product.
It may belong to the AI that decides which product gets bought in the first place.
