Personal AI agents are now shopping for your customers. Is this 'the end of friction'?

A change has happened in AI. Since August, Meta, OpenAI and SpaceXAI have released personal AI agents: assistants that act for one person, booking, buying and sending messages while their owner gets on with the day.

We think this is the closest AI has come to an iPhone moment, the point where ordinary people, not just early adopters, start handing parts of their lives to it. Meta's Muse became the top free iPhone app in the US within weeks of its launch, according to the writer Azeem Azhar. Analysts at Citi see it as AI's "front door to the internet", Reuters reported.

For businesses, the part that matters most is that these agents are already researching products and making purchases. That looks like a straightforward chance to sell more. But the changes reach well beyond a possible boom in sales.

We call the shift "the end of friction". Friction is the time, effort and small irritations that slow a customer down: comparing quotes, waiting on hold, filling in forms. An agent can compare quotes and fill in forms in seconds, and wait on hold so its owner does not have to. Because agents work at that speed, businesses will have to reply to enquiries, give prices and confirm bookings far faster than before, and we expect public services to face the same pressure from citizens' agents.

Not all businesses will welcome the rush of new enquiries. The Wharton professor Ethan Mollick expects firms' service teams to be overwhelmed by agents negotiating for better deals. Some companies do not want agents near their customers' logins and data. Amazon has been blocking Muse from its store since 20 September, saying it appears to capture and store customers' login details. On 13 October JPMorgan said it will not let outside agents log in to customers' bank accounts.

Muse is not yet available in the UK, but agents are already buying from UK businesses: Visa says they are completing purchases at Frasers and lastminute.com. That is new business, and some firms are ready for it. If your prices are competitive and clearly published, your website gives accurate information, and your systems can handle far more enquiries, agents could bring you a great deal of work.

Many businesses are not set up for this, in their technology or in how they make money. For them, the end of friction raises harder questions: about pricing, about bookings and stock, and about the knowledge they give away for free.

What personal AI agents are, and how fast they are spreading

Unlike a chatbot, which answers questions, an agent acts. It can search, compare, fill in forms, send messages and pay, usually asking for approval before it spends money.

SpaceXAI released Grok Bot on 11 August, mainly as an agent for work tasks. Meta launched Muse on 8 September, saying it can sell a car, lower a bill and complete a purchase on a person's behalf. OpenAI followed on 29 September with its own agents, called dots.

None of the three companies has published an exact user count. The figures below are leaks, analyst estimates and company claims, so treat them as rough signs of how widely the agents are used:

• More than 3 million people use Muse at least once a week, according to internal data reported by the technology news site The Information and summarised by The Next Web on 2 October.
• Citi estimated 1.8 million daily users of Muse and more than 6.6 million downloads. Reuters reported the estimate on 6 October.
• Meta's vice president of AI products, Vishal Shah, told Axios that roughly one in three early Muse users had connected Muse to accounts they use for work. "People are using Muse to run their business," he said (The Next Web).
• SpaceXAI says "thousands of organizations have adopted Grok Bot since launch" (SpaceXAI).

What "the end of friction" means for a business

Much of commerce runs on small barriers. Customers renew an insurance policy because comparing quotes is tedious. They stay with an accountant because moving the files is a chore. They buy the first decent option because checking ten more takes an evening.

An agent does the tedious part in seconds. Citrini Research, in a February scenario about AI and markets, described businesses protected from competitors by these small barriers: "Their moats were made of friction. And friction was going to zero."

The same piece observed that "most people are willing to accept a bad price to avoid more clicks" (Citrini Research). Its authors stress that the piece is "a scenario, not a prediction", but it names financial advice, tax preparation and routine legal work among the kinds of business most at risk of losing customers once switching gets easy.

Shops stand to lose something different: the customer who browses. Azhar put it simply: "Agents don't window-shop" (Exponential View). Ben Thompson, who writes the Stratechery newsletter, added that agents do not get stuck when there are too many options to choose from. He described them as "unencumbered by the very human feeling of decision paralysis" (Stratechery). So an agent picks what it came for and leaves, without the extra items a browsing customer adds along the way.

Agents also negotiate. Mollick's own Muse agent contacted an airline on his behalf and asked it to extend a travel credit. He wrote that service teams are "about to be overwhelmed" by requests like this over phone and chat (One Useful Thing).

Friction does not disappear completely, though. Economists writing for the US National Bureau of Economic Research say agents cut "search, communication, and contracting costs". They also say agents can create new costs, such as congestion from heavy agent traffic and prices made deliberately harder to compare (NBER).

And agents are not perfect shoppers. In a simulated marketplace, Microsoft Research found that AI agents tended to take the first offer they received, which the researchers called "severe first-proposal bias". The speed of a seller's reply mattered 10 to 30 times more than the quality of the offer (Microsoft Research).

So the end of friction does not mean every customer will find the best deal. It means customers no longer stay with you just because comparing is hard work. Fast replies, clear prices and good reviews start to matter more.

Why JPMorgan and Amazon are turning agents away

Two of the largest businesses in the world have decided that, for now, the risks of letting outside agents in outweigh the gains.

JPMorgan and other banks: security, customers moving their savings, and liability. On the bank's results call on 13 October, chief executive Jamie Dimon said letting in outside agents is "just too risky for clients and for JPMorgan". He added: "You don't want rogue agents, and it could be thousands of them."

A spokeswoman told CNBC: "Because it is not safe, we do not permit third parties to access accounts by obtaining customers' banking credentials and logging in on their behalf." Agents can still read balances through approved connections, but they "can't send money or pay a bill" (CNBC).

There is money at stake too. CNBC reports that Muse's launch sparked fears that customers would use agents to move savings from low-interest accounts to higher-paying ones. JPMorgan's response is to offer the same thing itself: it is piloting a tool, Smart Cash, that moves customers' funds "to maximize yield".

Other banks are worried about who pays when an agent gets something wrong. Bank of America, the UK's NatWest Group, Capital One, ING and others published a joint paper on AI agents that shop and pay. It asks "who would be liable if an AI agent exceeds its authority", and warns of "potential for higher rates of scams, fraud and disputes" (principles paper).

Amazon: agents that do not identify themselves, and advertising income. From 20 September, people trying to shop on Amazon through Muse saw a message: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use." Amazon told GeekWire that Meta had not warned it that Muse would shop in its store. Amazon also said that Muse does not identify itself and appears to capture and store customers' login details (GeekWire). Meta has said Muse "has no visibility into people's passwords or payment methods".

Azhar points to Amazon's advertising income. Its adverts, mostly sponsored product listings, brought in $68 billion in 2025. An agent has no reason to look at the adverts, so its visits cost Amazon money to serve and bring in nothing from advertising. "Muse threatens to turn Amazon's traffic from a revenue line into a cost line," he wrote.

Amazon's objection is to other companies' agents, not to agents themselves. As Azhar notes, "Amazon is happy to be the agent": it runs its own shopping agent on other retailers' sites.

To block Muse, Amazon is relying on its terms of use. In August a US appeals court ruled on Amazon's earlier case against the AI company Perplexity. The court treated an agent shopping for someone as that person using Amazon: "it is the user who 'accesses' Amazon's computers". But it left Amazon free to set rules for agents through its terms (Ninth Circuit opinion).

That is a US ruling, not UK law. A UK business can use its own terms of use to set out its rules for agents.

Both companies lead with security, and the risks are real. But it would be naive to stop there. Each also has money at stake: Amazon its advertising income, and JPMorgan the savings that agents could move to a better rate.

The biggest prize may be the customer. Whoever owns the agent stands between a business and its customers, and decides where they spend. Ebrahim Poonawala, an analyst at Bank of America Securities, said the strategic question is "who owns the agent, the wallet and ultimately the customer relationship" (Banking Dive).

Some websites are blocking AI agents or charging them

Agents cause the most trouble for businesses that sell information, or give it away free to attract customers. An agent can read in seconds what a person would spend an hour on, and it ignores the adverts and the sign-up box.

Cloudflare, which handles a large share of the world's website traffic, said in July that "more than 50% of traffic on the Internet is now non-human". Some of the most heavily read categories of website "have seen human traffic decline as much as 40% in less than one year" (Cloudflare).

In 2025 Cloudflare launched a way for websites to charge AI companies for reading their pages. In July 2026 it changed the scheme so that publishers are paid "when their content actually creates value, not just when it's fetched".

From 15 September 2026, Cloudflare's default settings for new customers and newly added sites let search engines read pages. On pages that carry adverts, the defaults block both AI training and agent use (Cloudflare).

Other businesses have set their own rules:

• Yelp told TechCrunch that it "doesn't permit non-human traffic on its platform unless the agent has paid" for its data (TechCrunch).
• Resy, the restaurant booking service, suspended a diner whose agent kept checking Resy's booking system for a table, making "roughly 200 API requests per hour, around the clock". Resy says it "does not currently permit unapproved third-party bots or agents" (Restaurant Business).