What happens erstwhile we go buying for artificial intelligence alternatively of looking at offers at the store? Meta and Amazon are fighting for influence on our decisions, and with it for gross from advertising and transactions. We're explaining who can make money on the improvement of AI agents and what investors should pay attention to.
In September 2026, Meta (Facebook) launched Muse – an AI agent who can browse the pages on his own, find products, compare offers and go through subsequent acquisition stages. The user only has to approve the transaction at the end.
In just 12 days, the app was downloaded to iPhones in the USA and Canada 1.8 million times. For comparison, ChatGPT recorded about 1.3 million downloads in the same period after his debut.
Shortly after that, Amazon cut Muse off from his store. The question is: why? From a customer's point of view, this solution seems very convenient.
The problem is that if an agent starts choosing products before the user enters Amazon or the site of a circumstantial store, these platforms will lose part of the impact on his decision. And then it can besides change who earns from choosing a given product alternatively of another.
Amazon closes the door to Muse
Muse doesn't work like a common chatbot. It receives a task and can open websites, search for information and follow up on behalf of the user.
The difference is well seen in a simple example. Chatbot can tell you which hotel to choose. Muse can open the pages himself, fill out the forms and work on the task even after the application is closed. It returns to the user erstwhile his decision is needed.
Graphics generated by AI
But Amazon doesn't want Agent Meta to do it in his store.
The company first asked Meta to exclude Amazon.com from Muse. Meta said no, so Amazon started blocking an agent's access. A user attempting to enter the Amazon website via Muse received a message that the usage of an unauthorized AI agent violates the rules of service.
Amazon explains the decision, among others, about the authorisation of the agent, its recognition and the safety of client data. The company claims that Meta did not get approval to operate Muse in the store, and the agent can access the client account pages and the past of his orders.
The meta presents the safeguards differently. The company claims that the authentication data is stored separately, and the agent alone does not have access to passwords or payment data.
The dispute isn't just about Muse. Amazon had previously sued Perplexity in connection with the operation of his purchasing agent, as well as limiting the access of OpenAI and Google agents.
At the same time, Amazon develops its own solutions with a very akin effect.
In 2025 Amazon launched the Buy for Me. If the user finds a product in the Amazon app that Amazon does not sale itself, the agent can go to the outside of the brand and acquisition there on its behalf.
A company that doesn't let another people's agents in, so it sends its own agent to another people's stores.
The difference is important. In the case of Buy for Me, the client continues to acquisition in Amazon. It is Amazon who helps to find the product and decides erstwhile to direct the user further.
Muse works backwards. The user starts with Agent Meta, and Amazon becomes 1 of many stores for him to check.
Agent picks the product before the client enters the store
Imagine individual wanting to buy a laptop up to PLN 5000.
Today it can kind specified an enquiry into Google, enter Amazon or open pages of respective electronics stores. It reviews results, reads reviews, compares processors, memory, screen, or battery time.
On the way he sees advertisements, sponsored results and recommended products. any of them were paid for.
Agent AI can be commissioned in 1 sentence:
"Find me the best laptop to 5000 zł for work and light movie editing".
The agent can review the shops and reviews himself, choice respective models, check prices and prepare the purchase. The user is only needed to approve the transaction.
So the product was selected before the client entered the website of a peculiar store.
It's a large difference for Amazon. If you first ask Muse, the agent can choose a circumstantial laptop and then direct it to the store where you can buy it. The client does not then enter on the Amazon "laptop to PLN 5000". He's coming in for a circumstantial model.
Amazon can inactive sale the product, handle the payment and deliver the package. However, it loses any of the influence on the earlier phase erstwhile the user is just wondering what to choose.
And that's erstwhile product ads are peculiarly valuable.
Continue reading: ]]>Meta versus Amazon. Who gets paid erstwhile AI starts buying for us?]]>
]]>How do news influence investor performance?]]>
Author: ]]>Piotr Chuszno]]>
