As artificial intelligence (AI) agents gradually have the ability to independently search, shop, and complete transactions, traditional internet platforms are facing an important choice. Should they actively embrace third-party AI shopping assistants or limit their access to protect their own user relationships and business models? Bernstein pointed out in the latest research report released on October 23 that for Internet platforms with high market shares such as Amazon (AMZN.US), DoorDash (DASH.US), and Uber (UBER.US), denying or restricting third-party AI agent access at this stage may be a completely reasonable business strategy.
The Zhitong Finance App learned that Bernstein analysts Nikhil Devnani and Mark Shmulik believe that although AI agents are expected to completely change the way consumers search, discover products, and complete transactions online in the long run, the market's current concerns about disrupting the business model of traditional internet platforms may be excessive. AI agents are still in a very early stage of development. Technical capabilities, consumer usage habits, and business cooperation models are not yet mature. There is no need for existing platforms to rush to help potential competitors establish new trading portals.
The research report points out that the key to whether consumers are ultimately willing to switch from traditional e-commerce and service platforms to AI agents is not the technical performance of the AI model, but rather whether these products can provide a shopping experience that is clearly superior to existing methods.
What consumers really care about is price, trust, convenience, difficulty of operation, and whether tasks can be completed accurately and in a timely manner. For consumers who are already accustomed to using mature platforms such as Amazon, AI shopping assistants must not only complete the same tasks, but also show sufficient obvious advantages in terms of reliability and convenience in order to encourage users to change their long-term consumption habits.
Bernstein emphasized that after years or even decades of development, existing Internet platforms have established mature brand reputation, product search systems, transaction processes, and user service capabilities. Taking Amazon as an example, the process for consumers to search and buy everyday necessities on its platform is highly mature, and it is not easy for third-party AI agents to replace this experience.
At the same time, there are also obvious differences in the difficulty of applying AI agents in different consumption scenarios.
The bank summarizes the main uses of consumer AI into three categories: completing tasks for users, purchasing products for users, and providing entertainment services. Among them, shopping is considered to be one of the most difficult areas to achieve large-scale behavioral change.
Bernstein pointed out that even though e-commerce has been developing for about 30 years, its penetration rate in the US overall retail market is still only about 20%, and the penetration rate in the global market is even lower. This shows that changes in consumer shopping habits usually take a long time, and the advent of AI technology does not mean that all transactions will quickly switch to new channels.
However, the early performance of Muse, an AI agent under Meta (META.US), has shown the potential market size for consumer AI shopping. According to data disclosed by Meta during the third quarter of 2026 earnings call, Muse attracted 2.5 million daily active users in the North American market after only about a month and a half of its launch. What's more noteworthy is that about 25% of Muse weekly active users have tried using AI agents to shop, making it the most popular application scenario for this product.
Meta also revealed that the average purchase amount of Muse users was over $200, including travel-related spending, and that some users on the company's platform with a high average income per user also became early Muse users.
Bernstein believes that these data suggest that AI proxy shopping is gaining some degree of consumer acceptance. However, it is still difficult to determine whether early trials can be converted into long-term stable usage habits.
The bank specifically emphasized that user retention rate and actual usage scenarios will be an important indicator for measuring the long-term value of consumer AI products. Even if AI agents can attract a large number of users in a short period of time, it may be difficult to maintain user activity if they cannot continue to provide reliable and convenient services.
At the technical level, Bernstein believes that one of the key factors that determine whether AI agents can truly become popular is their ability to interoperate with external websites and service platforms.
Meta's Muse uses virtual machine technology to access and operate websites, enabling AI agents to perform tasks in a wider range of internet environments, and the integration of payment systems also helps enhance consumer trust.
However, there are still significant limitations to this approach. Analysts Bernstein discovered in actual experience that although Muse can help users quickly find the materials needed to apply for a passport, it has also encountered robot verification barriers on websites such as Yelp and Wayfair (W.US), making it impossible to successfully complete the task.
For ordinary consumers, such issues may directly affect their trust in AI agents. The bank pointed out that shopping is different from general information inquiries. Once an AI agent selects the wrong product, misses an order, or is unable to complete the transaction during the purchase process, consumers will not only not be able to save time, but may also require additional processing of the problem. Therefore, official technical interfaces and commercial partnerships are critical to the development of AI agents.
Compared to directly operating a website designed for human users, completing product inquiries, order processing, and payment through an application programming interface (API) can generally provide a more efficient and reliable transaction experience. Currently, Muse has established partnerships with companies such as Shopify (SHOP.US), Walmart (WMT.US), Instacart (CART.US), Expedia (EXPE.US), and Wayfair (W.US).
Bernstein believes that for companies that have a relatively low market share and want to challenge industry leaders, partnering with AI agents may be clearly appealing. For example, Walmart competes with Amazon in the e-commerce field. By connecting to third-party AI shopping channels, Walmart has the opportunity to obtain additional traffic, expand market share, and increase revenue.
However, for platforms such as Amazon that already have a huge user base and mature trading ecosystem, the situation is different. Bernstein pointed out that Amazon has now adopted a strategy to restrict third-party AI agent access while actively developing its own Alexa shopping function.
The business logic behind this choice is that once a third-party AI agent becomes the main entry point for consumers to shop, traditional e-commerce platforms may gradually lose control over user relationships, product discovery processes, and transaction decisions.
At that time, consumers may no longer directly access platforms such as Amazon, but use AI agents to compare products, select merchants, and complete purchases. Traditional platforms may be weakened to the infrastructure that provides products, fulfillment, and other back-office services.
This is exactly the risk of “disintermediation” that internet platforms are concerned about. Bernstein believes that for companies that occupy an important position in their respective markets, such as Amazon, DoorDash, and Uber, cooperating with third-party AI agents too early may be tantamount to actively helping future competitors establish user portals. In particular, it is often difficult for third-party AI agents to bypass platforms that have a unique supply of products or services and large amounts of offline resources to complete transactions directly.
This supply-side advantage allows existing platforms to still have some initiative when negotiating with AI agent developers. Therefore, Bernstein believes that these companies can choose to wait and see for a while, and then decide whether to open up cooperation after the scale and business model of consumer AI agents are more clear.
This strategy can not only slow down potential risks of disintermediation, but may also buy time for the two sides to establish a more favorable commercial cooperation mechanism in the future. However, Bernstein also warned that rejecting third-party AI agents does not mean that existing platforms can permanently withstand technological changes.
If consumers eventually switch to AI agents on a large scale, and external AI shopping channels begin to contribute a significant proportion of an industry's new transaction volume, then it may be difficult even for market leaders to continue to ignore this channel.
For industries where the supply of goods and services such as travel is already highly digitized, AI agents may be easier to directly connect with airlines, hotels, and other service providers, thereby bypassing traditional online travel platforms.
In contrast, in fields that are highly dependent on offline merchant networks, delivery capabilities, and platform services, such as takeout delivery and instant retail, existing aggregation platforms may have stronger defense capabilities.
Bernstein also pointed out that one of the important competition in the future consumer AI market will occur between general-purpose AI agents and AI agents dedicated to vertical fields. General-purpose AI agents seek to become a unified portal for consumers to handle various online and offline transactions, while platforms such as Amazon can use their own user purchase history, payment information, membership relationships, and logistics services to develop AI shopping tools more suitable for specific scenarios.
The bank believes that whether Amazon can successfully promote the Alexa shopping function will be an important indicator for observing this competition. If even Amazon, which has a huge user base and a mature e-commerce ecosystem, cannot promote the widespread use of its own AI shopping agents by consumers, then smaller Internet platforms may face greater challenges if they want to establish successful vertical AI agents.
From an investment perspective, Bernstein currently maintains the “outperforming market” ratings of Amazon, Meta, DoorDash, Uber, Instacart, and Wayfair. Amazon is given a target price of $320, Meta's target price is $850, DoorDash's target price is $270, Uber's target price is $95, Instacart's target price is $60, and Wayfair's target price is $125. These target prices and ratings are part of Bernstein's current coverage and are not new adjustments made by the bank as a result of this AI proxy study.
Overall, Bernstein believes that AI agents may have a profound long-term impact on the Internet business model, but this change will not happen overnight in the short term. At a stage where consumer usage habits, technical reliability, and platform cooperation mechanisms are not yet mature, Internet platforms with a high market share choose to restrict third-party AI agent access, which does not mean refusing to innovate, but may be a reasonable choice to protect their own commercial advantages and seek time for strategic adjustments.
However, the bank also emphasized that in the end, it will still be consumers who determine the industry pattern. If AI agents can continue to provide a shopping experience that is clearly superior to traditional platforms and gradually become the preferred transaction channel for consumers, then existing internet giants must also adapt to this trend, otherwise they will face the risk of losing user entry.