We need to get used to AI agents. Instead of simply answering questions, agents are programs that perform multi-step tasks in a chain, such as searching for a product, comparing prices, and completing the purchase. However, their biggest obstacle is no longer technical: it’s the shops that decide whether or not to let them in.
Amazon hasn’t decided, but Walmart has (albeit on its own terms).
In short. It was Walmart’s CEO himself, who presented the strategy during an event in San Francisco to allow agents to work on behalf of users at both Walmart and Sam’s Club subsidiaries. It is about the following:
- Agent Gateway– A server based on Anthropic’s model context protocol for agents to connect to external data and tools. Allows authorized agents to search online inventory and make purchases on the user’s behalf. This is something that is already starting as a beta for both the company and the developers.
- Agent Profile: This is coming next month and that’s when customers can activate their agent and link it to their Walmart account. Once activated, you must set certain controls and preferences, such as spending limits Additionally, the customer can delete the agent profile at any time.
Conditions. These features will be available to any Walmart account holder, but agents won’t be able to roam wherever they want. Presumably, they will not be able to go beyond what humans establish or access specific information: they will only gather what is strictly necessary to complete their tasks.
Also, and here may be a concern, Walmart will allow the return of products that ‘unregulated’ agents try to purchase. Agents who fail to comply with these rules will be blocked… and only those authorized will be able to make purchases.
Another interesting thing about gateways is that these chatgpt, gemini, muse or authorized agents can connect directly to the company’s system, but the prices they see are the same as what a person would see. It’s… trivial, but it’s easy to imagine that there could be agreements between companies to promote the use of agents to promote technology. It seems that will not be the case.
At the far end. Thus, Walmart adds external agents to the purchase assistant they already had (Sparky), but on the opposite side we find the main competitor: Amazon. Meta launched Muse in early September, but it didn’t take long for Amazon to issue a notice that access to an unauthorized AI agent would violate its terms of use.
In fact, Amazon impersonated Muse, saying they did not authorize Meta Agents to access customer accounts, extract data, or process transactions. As we read in the implicator, they add that the meta did not inform them and the agent does not identify itself while browsing. This is not the first time, as they have restricted the agents of Google, OpenAI and even sued the confusion for the Comet browser.
Agent Jam. Here’s another interesting read, and that is that if an agent makes a mistake and orders an unauthorized item on Amazon, it’s Amazon who eats the processing and return costs, not the company responsible for the agent. Additionally, an agent won’t buy on impulse, something a human does, so there, stores like Amazon might lose a bite. And, on the other hand, it is logical to think that they want to promote their own agents and assistants.
There are several themes. On the one hand, big companies don’t want to hand over the client relationship to an outside agent. On the other hand, some of the companies with AI chips, servers, models and infrastructure run the biggest stores on the planet. Alibaba and Amazon are two examples, and if they build agents, it’s logical that they want their systems to work in their stores.
The problem for the user is, if there is no value, then we are heading for a future where if we used to think of paying an agent to manage the whole digital ecosystem, now it seems we have to have one person per store because everyone wants their own or, if they don’t develop it, they can veto someone and not others.
What is at stake?. For example, the business model of companies like Instinct depends on giving access to it to external companies, and recently the head of Mater’s artificial intelligence laboratory noted that the company’s finance director told him Muse had to work no matter what. Why are agents interested in work? Well, because Meta itself has given Muse a key to monetization.
They estimate that 10 million users use it and one of their monetization plans is to keep a percentage of the money from the items purchased by the customer. It will be like an affiliate system: it won’t cost the client much, but Meta will ‘suck’ from every coffee maker that its agents buy on Amazon. And Amazon doesn’t see a dime in it unless they reach a deal with Meta.
Picture | Sven
In Xataka | AI agents are learning to improve their own software. The key is to predict your mistakes before you make them.
