When Meta Muse Collides with Amazon: AI Is Leaving the Chatbox to Contest the New Internet Gateway

Recently, while talking with software and hardware entrepreneurs in Las Vegas and Silicon Valley, one change has become increasingly obvious. What entrepreneurs showcase is rarely "my model answers better than others" anymore; instead it is how AI actually enters the workflow: one Agent writes code, one reviews it, one searches for information, and another handles testing; human work gradually shifts from operating software by hand to assigning tasks to a group of digital employees, checking results, and handling exceptions.

Last month, Meta launched a personal AI Agent—Muse—pushing this change toward ordinary consumers. Muse has an independent virtual computer and browser, and can send emails, book travel, shop, and complete tasks on behalf of users. About two weeks after launch, Muse downloads had reached roughly 2.8 million. At the end of September, Meta pushed Muse further into the small-business market, connecting with tools such as Shopify, QuickBooks, Stripe, and Canva. On the October 21 earnings call, the company announced that more than 10 million people had activated Muse. In October, Wall Street had already begun to revalue Meta—as the "AI front door."

From competing for attention to capturing "real intent"

Over the past two decades, the starting point of internet business was the input box. But today in Silicon Valley, the input box itself is beginning to disappear.

Silicon Valley startup Typeless simply proposed that "the keyboard is a mistake." It wants users to speak directly, with AI turning unorganized speech in real time into sendable emails, messages, and instructions; what it is competing for is not the chatbot, but the keyboard and input box themselves. As Yu Meng, CEO of Shandian Shuo, sees it, the disappearance of the dialog box will likely begin with voice communication as a substitute—for the same meaning, when facing a customer, colleague, or friend, AI will automatically switch expressions according to the communication target and context.

In San Francisco, another group of entrepreneurs is trying to record meetings, consultations, on-site visits, and face-to-face conversations that happen outside the screen, following the path of "Capture, Remember, Act" to directly generate emails, CRM updates, and next-step tasks.

This happens to hit a strange blind spot of current large models. It can read the internet, yet it does not know which complaint a customer voiced that a salesperson heard ten minutes ago; it can read medical papers, yet it does not know what a doctor just discussed with a patient; it can summarize all the documents a company has publicly released, yet it does not know what decision management just made in the meeting room.

Weida Tan, founder of Silicon Valley company PressClub, told the author that this type of product most easily generates cash flow, because what it replaces is not ordinary operations but the hourly-billed work of high earners: "Lawyers and doctors in the United States are very willing to pay for this; although China has not yet formed a habit of paying for such services, in the United States this kind of subscription has already become a natural commercial loop."

It is precisely the payment habits of high-net-worth individuals that have enabled Silicon Valley entrepreneurs to quickly complete the positive commercial cycle from "capturing offline intent" to "triggering workflows." AI is retreating from an App that needs to be specially opened to the background layer of communication between people and computers—in the future, users may not even realize they are "using AI."

In the search era, Google competed for keywords. In the mobile internet era, Apple, Meta, and various Apps competed for clicks on the screen. In the Agent era, what is contested has become an entrance earlier than clicks and closer to people: before users open any specific website, who is the first to know people's real intent.

Meta's description of Muse is very direct: it does not just give suggestions, but "actually does the work." Muse runs in an independent Secure VM, has its own browser, and can, with authorization, access the internet and different services on behalf of users.

But less than two weeks after Muse launched, it clashed with Amazon.

Amazon blocked Muse from accessing its shopping site and even disallowed it from automatically browsing products. Amazon said AI Agents should clearly identify themselves and respect the service provider's decision on whether to allow Agents to enter. Previously, Amazon had also blocked shopping bots from Google and OpenAI, and had a legal conflict with Perplexity over similar issues.

On the surface, this is a website anti-crawler issue. In reality, what is being fought over is the most valuable layer of e-commerce: after consumers generate purchase intent, who controls everything that happens next?

In the past, a consumer who wanted to buy a pair of running shoes would open Amazon. He entered keywords, and Amazon decided the ranking of search results; brands bought ads to compete for higher positions; users read reviews, compared prices, and then completed payment on Amazon. The search entrance, product ranking, ad exposure, consumption data, and checkout were all within one system.

Agents have changed this path. In the future, consumers may no longer open Amazon, but instead say to Muse: "Help me buy a pair of running shoes for running 5 kilometers every day, under 200 dollars, not too soft." Next, which brands to search, which products to compare, which options to exclude, and even where to ultimately buy may all be completed by the Agent. Until the final transaction, the consumer may never have opened Amazon from beginning to end.

This is what Amazon truly needs to defend. What Amazon is blocking is not a bot, but a new internet entrance taking away consumers' intent.

This worry is no longer merely theoretical. In early October, Leif, co-CEO of U.S. investment platform Public In an interview, Abraham said that after using Muse, he began increasingly buying skincare products, jeans and other goods through Muse, and eventually many orders flowed to Shopify independent sites rather than Amazon. In other words, Muse has already begun to change the shopping paths of some consumers. What Amazon truly loses is not just a single order, but the entire decision chain from "what the user wants to buy" to "where they ultimately buy it."

"I now use instinct; I don't even open the Muse app. I just text instinct, and it books rooms and shops for me. It's more convenient," the Silicon Valley entrepreneur said.

If Muse still needs to reach into other people's internet, Google takes a different path: it already owns a large part of the real-world index itself.

Google launched Universal Cart this year. Users can add products to the same shopping cart through Search, Gemini and even Gmail; Gemini monitors prices, compatibility and inventory in the background, and then completes the purchase through Google Pay or the merchant's website. Google says its Shopping Graph already contains more than 60 billion product listings. The new version of Ask Maps launched in August can also assist with real-world tasks such as ordering food.

Search knows where information is, Maps knows where merchants are, Gmail knows what users are working on, Chrome knows how to enter websites, Android connects devices, and Google Wallet handles payments.

In this battle to defend the entry point, Muse tries to take over operations across platforms with a virtual browser, while Google tries to use its own native ecosystem to keep every step from intent to execution within its own yard.

On the day Muse was released, Meta partnered with Stripe to integrate it into Link. After U.S. users authorize it, Muse can use the payment methods users have already saved to complete checkout at more than 1 million merchants that accept Link; for other merchants, Link can also issue the Agent a one-time virtual card used only for approved transactions (sensitive operations such as purchasing goods still require user confirmation).

As a result, a new transaction chain has emerged on the internet: human intent → AI Agent → comparison and decision-making → merchant → payment.

In the past, the chain Amazon built was: human intent → Amazon search → Amazon ranking → Amazon ads → Amazon payment.

The two are colliding head-on. This is not competition between two products, Muse and Amazon, but competition between two generations of internet distribution models.

When payment tools such as Stripe help AI complete the final link of transactions, the real crisis for traditional e-commerce platforms emerges: if the Agent is the first to know the user's intent and can also search, compare, decide and pay on the user's behalf, then what Amazon is blocking is not Muse, but a future in which Amazon may be downgraded from a "shopping mall" to a "shelf."

Wall Street has already begun pricing Muse according to this logic. On October 6, a Citigroup analyst called Muse a possible "front door to the internet" in the AI era, estimating that by 2030 Muse could contribute more than $27 billion in revenue to Meta. This valuation logic is worth noting: what the market values is no longer just whether Muse can charge fees, but that if it becomes the first entry point for users to search, compare and transact, Meta may migrate from being an "attention platform" in the past to an "action distribution platform."

What Amazon, Google and Meta are fighting over today is no longer just models, nor just traffic. They are fighting over a more fundamental question: who has the power to act on behalf of users.

Three years ago, the most striking symbol of the AI revolution was a blinking chat box in a browser. Now, that chat box is slowly receding into the background, and AI is beginning to enter meeting rooms, glasses, input boxes, browsers, maps and wallets, quietly embedding itself into dozens of real-world actions people take every day.

Over the past three years, the problem Silicon Valley solved was how to give machines an increasingly smart brain. The bigger business question of the next stage may be: who is the first to know human intent? Who has the right to act on people's behalf? And finally the most practical question: when AI starts making decisions for you, who does it ultimately spend the money with? And who will bear responsibility for AI's consumption decisions?

(The author is the founder of Silicon Valley YuanTiao Coremi. Contact email: yuanlu007@pku.org.cn; disclosure of interests: the author's company holds a Coremi index portfolio of tech stocks including Google and Nvidia. This article represents only the author's views.)