Think about how we use Amazon.
Amazon provides the marketplace, but it also controls the interface through which customers discover products, compare sellers, read reviews and make purchases.
Now imagine asking an independent shopping agent to do all of that on your behalf.
You describe what you're looking for, explain your preferences and let the agent research available options. It compares products, evaluates alternatives and presents a recommendation. You approve the purchase, and Amazon handles the order and fulfilment.
The service hasn't disappeared. Amazon is still providing it.
But you didn't need to interact with Amazon to access it.
What happens when the company providing a service no longer controls the interface through which its customers interact with that service?
Who controls the interaction?
There are a few ways this could play out.
1. The platform controls the interaction
Amazon already offers its own shopping assistant. In India, Rufus was introduced as a conversational shopping assistant integrated into the Amazon shopping experience. Customers can ask questions, compare products and discover options without leaving Amazon. Amazon has since brought Rufus and Alexa together under its Alexa for Shopping experience in the US. [1] [2]
The interface becomes conversational, but Amazon retains control over the experience.
It still has access to customer preferences, purchasing history and the surrounding commercial environment. It can recommend products, introduce alternatives, suggest complementary items and influence where the customer spends their money.
We are seeing similar experiments outside retail. MakeMyTrip, for example, has developed Myra, an AI-powered travel assistant built on an agentic AI framework. It can take travellers from initial search to confirmed, paid booking within a single conversational interface, including through voice interactions. [3]
The platform is changing how customers interact with its services, but it isn't giving up control of that interaction.
2. A third party controls the interaction
Now imagine using an agent provided by another company.
Instead of opening Amazon, you ask an independent agent to find something for you. It might search Amazon, compare products across other marketplaces and recommend where to buy.
Amazon is now one of the services the agent can access.
The company providing the agent controls the interaction and potentially influences what the customer discovers, which products get recommended and where the transaction happens.
It might also gain visibility into the alternatives customers considered, the preferences they expressed and the reasons behind their decisions. Amazon might still know what the customer eventually purchased, but it might no longer have the same visibility into the decision that led to that purchase.
3. The user controls the interaction
A third possibility is that the user brings their own agent.
They might configure it, build it or choose the models and services it uses. They decide which sources it can access and what preferences it should follow.
The agent could compare products across multiple platforms and select whichever best meets the user's requirements.
In this arrangement, preferences and decision history could remain with the user and be carried across different service providers, rather than being confined to a particular platform or agent provider.
These aren't necessarily successive stages. All three arrangements could coexist.
The difference is who controls the agent, whose interests it is expected to serve and who gets visibility into the customer's decisions.
Who captures the economic value?
This is where the interests start conflicting.
Platforms such as Amazon have built their businesses around more than simply facilitating transactions.
They also operate advertising networks, collect information about purchasing behaviour, provide seller services and create opportunities for cross-selling.
An external agent could take over some of the activities that generate those opportunities.
The agent provider might introduce its own commercial arrangements. Sponsored recommendations, affiliate commissions and promotional placements could influence which products or platforms get recommended.
Amazon operates a marketplace where independent sellers compete, but it also competes through its own private-label brands, including Amazon Basics, Solimo, Presto!, Vedaka, Symbol and Myx. [4]
This gives Amazon a commercial interest in retaining influence over product discovery and recommendations.
An independent agent could steer a purchase towards another marketplace or seller, shifting some of that influence elsewhere. The same commercial incentives could emerge at another level.
A user-controlled agent could operate differently, prioritising the user's preferences rather than the commercial objectives of a particular platform. That doesn't automatically make it unbiased, but it changes who gets to configure its priorities.
The transaction itself is only one part of the economic value. Access to customer attention, preferences and purchasing decisions matters too.
What happens when platforms restrict external agents?
Existing platforms have already demonstrated that they want control over how external applications access their services.
Reddit changed its API pricing and access terms in 2023. The changes made it financially unsustainable for some commercial third-party applications. Apollo, one of the popular Reddit clients, shut down in June that year. [5] [6]
Twitter, later renamed X, made its own TweetDeck product, renamed X Pro, available only to subscribers in 2023. [7]
These weren't agent-related disputes, but they illustrate something important: platforms can change access policies when alternative interfaces conflict with their commercial interests.
We're now seeing similar tensions emerge with agents.
In September 2026, Amazon blocked Meta's Muse from shopping on its website. Amazon said third-party purchasing agents should operate openly and respect service providers' decisions about whether to participate. [8]
This is an interesting example because Amazon isn't simply another product catalogue. It controls the service that the external agent wants to interact with.
There is also a practical concern. If an external agent misunderstands a request or places the wrong order, the customer and seller might still turn to the platform to resolve the problem, even though the platform didn't control the agent.
Platforms don't necessarily have to accept every external agent that wants access.
They can restrict access, introduce commercial arrangements, build their own assistants or selectively expose capabilities through APIs.
At the same time, not every platform is resisting external agents.
Interestingly, the same Amazon that blocked Meta's Muse has made parts of its advertising platform available to external agents through its MCP server. Introduced in February 2026, it allows connected agents to create, update and delete campaigns, retrieve reports and manage account settings. The integration is available to Amazon Ads partners with active API credentials. [9]
The distinction is worth noting. Amazon restricts an external agent from accessing its retail site without authorisation, while allowing external agents to operate within a defined advertising integration that Amazon controls. A platform might welcome agents for one set of interactions while restricting them from another.
Figma provides another example from outside retail. Its MCP server allows supported agents to read design files and write native content back to the canvas, including creating and modifying components, variables and screens. [10]
Salesforce is taking a similar approach with enterprise software. Through initiatives like Headless 360, it is making capabilities across its product suite available to authorised external agents through MCP servers and APIs. Its partnership with Anthropic, Claudeforce, takes this further by bringing Salesforce data, workflows and business capabilities directly into Claude, while retaining existing permissions and business rules. [11] [12]
These examples might point to an emerging pattern. Software platforms could be among the early adopters of this shift, exposing their existing capabilities to agents rather than requiring users to interact with their traditional interfaces. Whether businesses outside the software industry will also begin exposing their services to agents they don't control remains to be seen.
The intermediaries don't necessarily disappear
Making services accessible to agents doesn't automatically make every business discoverable.
An agent cannot reasonably evaluate every available product, seller and service every time a user asks for something.
Someone still has to decide where the agent should look and which information it should trust.
For independent businesses, discoverability might increasingly mean making their products and services accessible to agents, not just visible to people browsing storefronts.
Perhaps users will have more control over their preferred sources, similar to how people use RSS feeds to choose what they follow rather than relying entirely on a platform's recommendation algorithm.
Another possibility is that new intermediaries emerge around reputation and trust.
A service could help agents identify reliable businesses, verify information and resolve disputes when something goes wrong.
The internet already has a substantial trust deficit. An agent making decisions and transactions on behalf of someone else doesn't automatically solve that problem.
And if a new intermediary becomes responsible for establishing trust, it could eventually acquire its own gatekeeping power.
Removing one intermediary doesn't necessarily eliminate intermediaries. It might simply change where they sit in the relationship.
Where does this leave us?
The interface and the service don't necessarily have to belong to the same company.
Some platforms might build their own conversational assistants. Others might expose their capabilities to external agents while retaining control over how those capabilities are accessed.
Separating the interface from the service doesn't eliminate commercial interests or platform control. It changes where those interests sit.
Platforms want to retain the customer relationship, agent providers want to become the customer's preferred intermediary, users want agents that act in their interests, and service providers want their offerings to remain discoverable and accessible through whichever interfaces customers use.
Whoever controls the interface can shape the customer relationship and may eventually become a gatekeeper in its own right. Who eventually controls that interaction, and where the economic value accumulates, remains an open question.
References
1 Introducing Rufus, its conversational shopping assistant - Amazon India
2 About Alexa for Shopping - Amazon
3 MakeMyTrip Launches Myra, an AI-Powered Multilingual Travel Assistant to Simplify Trip Planning CXO digital pulse
4 Why India's small sellers still don't trust Amazon - Rest of World
5 Key Facts to Understanding Reddit's Recent API Updates - Reddit
6 Apollo is shutting down as a result of Reddit's new API pricing - TechCrunch
7 X makes X Pro (formerly TweetDeck) a subscriber-only product - TechCrunch
8 Amazon blocks Meta's Muse AI agent from shopping on its website - GeekWire
9 Introducing the Amazon Ads MCP Server - Amazon Ads
10 Get started with the Figma MCP server - Figma
11 Expanding Headless 360: Enterprise Capabilities - Salesforce
12 Salesforce and Anthropic Announce Claudeforce - Salesforce
Comments
Join the discussionNo comments yet. Be the first to comment.
