On September 8, Meta, the company behind Facebook, Instagram and WhatsApp, introduced a personal AI assistant called Muse. It is a program that does not just answer questions but gets things done. Muse runs on its own secured computer at Meta and, according to the announcement, “can open a browser, fill out forms, and negotiate on their behalf”. Longer tasks continue after you close the app. Before sensitive steps, such as sending an email or buying something, it asks for permission first. Payment runs through Link by Stripe, which creates card numbers for the assistant that work only once.
For now, Muse is only available in the United States. Still, the examples Meta gives itself say a lot about where this is going. Next to booking a trip and sending an email, Meta lists “selling a car for more” and “lowering a bill”.
Less than two weeks later, Amazon shut the assistant out of its store. According to Amazon, assistants like this should operate openly and respect a store's decision on whether to take part. Meta declined to remove Amazon from Muse, MediaPost reported. In early August, a US appeals court had already ruled in a similar case, about Perplexity's AI browser Comet, that it is the user who accesses a website, not the company that makes the assistant. The court saw the assistant as a tool working under that user's direction. That is American law, and the court deliberately kept the ruling narrow. But it shows how hard it is to treat an assistant differently from the customer who sends it.
The biggest effect is not in the online store
Most pieces about assistants like this are about shopping. Does the assistant find your product, can it check out, does it see your offer? That matters, but I think the biggest effect lies elsewhere. Not in the first purchase, but in everything that follows: the renewal, the subscription, the yearly price increase, the contract that quietly rolls over.
That is where businesses earn money from customers who do not look, and this is not a guess. The AFM, the Dutch financial markets regulator, published research in April 2025 into the premiums of property and casualty insurers, based on data covering 47.7 million policies. In 2023, almost half of the insurers studied charged a higher profit margin on at least one product to customers who had been with them for nine years or more than to customers of one to two years. A difference in risk did not explain it. For car insurance, this applied to 19 to 30 percent of insurers, for home contents insurance to 23 percent.
That works because almost nobody looks. Of the people with home contents or liability insurance, 88 percent had not checked in the past year whether they could get it cheaper elsewhere, according to the same report. For car insurance, the figure was 79 percent. The AFM also points to the British regulator, which showed that insurers built models to predict which customers would switch, in order to push their profit margin as high as possible.
So one side of the table already had a model. Now the other side is getting one too.
Insurers are simply the best measured example. Anywhere a contract renews by itself, a business can earn money in the same way from customers who do not look: subscriptions, maintenance contracts, business software, leasing.
What changes when the customer has an assistant
Comparing, cancelling and negotiating are chores people put off. They are dull, take time and only pay off later. I build AI assistants that take over recurring work myself, and postponed work with a clear goal is exactly where such software pays off most. As soon as part of your customers use an assistant like this, three things change.
- Prices that only hold because nobody looks will get looked at. Not once after an angry letter, but as often as the customer asks, at no effort to them. A quiet increase at renewal will stand out.
- Hassle no longer works as a barrier. Cancelling by registered letter, a long form, a price “on request”: for a person, those are barriers. An assistant fills in forms and keeps working when the app is closed. Whatever can fill in one quote request can send ten.
- What is not written down counts for less. An assistant compares what it can read: price, terms, delivery time, notice period. A good reputation that only exists in the heads of your regular customers does not automatically make it into that comparison.
The strongest counterargument
There are two good reasons to dismiss this as something for later.
The first: Muse only works in the United States, and a big store like Amazon simply shuts such assistants out. But a store of that size can afford to, because customers come anyway. An ordinary business that keeps out a customer's assistant mainly keeps out that customer. Nobody knows how quickly this will reach Europe. The building blocks are already there, though: Stripe has a payment method made for assistants like this, and Meta has the assistant itself.
The second reason is stronger. The same AFM report shows that most people who stay do so because they are satisfied with their insurer. Around three in four give that as their reason. People are loyal to their insurer, not to the policy itself, the AFM concludes. So loyalty is often real loyalty, and a piece of software will not make it disappear.
That is true, and it is why this is not bad news for every business. Satisfaction survives a comparison. A surcharge on loyalty does not. What an assistant removes is not knowing. A customer who is satisfied and then sees they pay more than a new customer for the same product will not stay satisfied for long, in my view. Businesses that already treat their regular customers fairly have little to fear. They even gain something: whoever has a sharp price now also gets found by customers who would never search themselves.
What you need to choose now
Hold your revenue up against one question: which part comes from customers who would pay less or leave if they looked carefully once a year? For that part, you have two options.
You can make the price defensible. Charge the same price to loyal and new customers, or reward loyalty visibly. Write down prices and terms so that an assistant can read them. Make cancelling as easy as signing up. That costs margin now, but you decide how much and when.
Or you leave it. Then that difference disappears as well, just not at a moment you choose: customer by customer, as soon as each one has an assistant take a look, and without you knowing who is already comparing.
Choose the first. You lose that slice of margin either way. Only with the first do you keep the customer.
Sources
- Introducing Muse: The World's First Personal AI Agent Built for Everyone, Meta, 2026-09-08
- Amazon Blocks Shoppers From Using New Meta Agent, MediaPost, 2026-09-21
- Ninth Circuit Narrows CFAA Reach in Perplexity Agentic Commerce Ruling, PYMNTS, 2026-08-06
- Een eerlijke premie voor loyale verzekerden, Autoriteit Financiële Markten (AFM), 2025-04-08