AI agents can already shop on your behalf, organize your digital routine, and even negotiate prices. The idea sounds super practical: you share your preferences, set a budget, and let the bot handle the rest.
But behind all that convenience, there is a point that still divides experts: is it really worth handing over purchasing decisions, along with your card details, to a system that is still in a very early stage of mass adoption?
Artificial intelligence technology applied to consumer spending, known as agentic commerce, is not truly mainstream yet. Even so, giants like American Express, Amazon, and Walmart are already racing to put this kind of feature in customers hands, seeing it as a new way to engage consumers and boost sales.
On the other hand, AI and security experts warn of a far less glamorous scenario: costly mistakes, poor decisions, and vulnerabilities that could expose your data to attacks. In other words, the same technology that promises to make your life easier can, if used carelessly, open the door to serious problems.
In this article, you will learn how these agents work today, what major companies are already offering, what the real risks are, and why, for many people in the field, it is still too early to give AI free rein to shop on its own for you.
AI agents and the whole agentic commerce thing
When we talk about AI agents in online shopping, we are not just talking about that basic chatbot that recommends products based on what you have already browsed. The idea is bolder: these are systems capable of acting on their own within certain boundaries, following instructions you define.
In the context of agentic commerce, the flow goes something like this:
- You set a maximum budget,
- explain your style or preferences,
- point to a store or a range of products you want,
- and the agent goes out, compares options, monitors the price, and can even complete the purchase.
This is exactly the kind of use case that has been catching the attention of major companies, even though the technology is still in a very experimental phase for mass consumer use.
What the big companies are doing right now
American Express and protections for AI-powered purchases
American Express recently announced new services and specific protections for customers who use certain AI agents to make purchases with their Amex card.
In practice, the company says it will be able to verify the identity of the agent when it makes a purchase, acting as an extra layer between the bot and the financial transaction. According to Amex’s own statement, the goal is to protect eligible customers against charges resulting from mistakes made by artificial intelligence agents.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
In other words, the company is already operating on the assumption that errors will happen and is positioning itself as a kind of safety net for anyone who decides to experiment with this type of shopping automation.
Amazon Rufus: the assistant that tracks prices and buys
Amazon created an AI assistant for its marketplace called Rufus. It is not a fully autonomous agent set loose in the wild, but it already gives a solid preview of where online retail is headed.
With Rufus, users can:
- ask questions about products,
- get help comparing options,
- track price changes,
- get notified when the price hits a target they set,
- and complete the purchase directly through Amazon’s interface.
Even though the final decision usually still goes through the user, the assistant already handles the heavy lifting: research, filtering, comparison, and timing the purchase just right. It is a concrete step toward an agent that, in the future, could operate with more autonomy.
Walmart and the conversational agent Sparky
Walmart, the largest retailer in the United States, also entered the game with an AI agent called Sparky. The company describes it as a conversational agent designed to:
- help consumers find products,
- show reviews from other customers,
- streamline the ordering process.
Sparky works as an assistant that accompanies the shopping journey, trying to make everything smoother and more personalized. Even without fully autonomous control over your card, it already acts as a smart intermediary between users and Walmart’s massive catalog.
Who is already using AI to research and buy
This wave is not just corporate talk. Data from a market survey conducted by Statista in November shows that, among Americans aged 18 to 39, roughly one quarter have already tried using AI tools to research products or make purchases.
So while usage is still far from standard, there is already a significant base of people testing this kind of feature in their everyday shopping.
Where things start to go wrong
As adoption grows, cases where things went very differently than expected are also starting to surface.
When the agent nails the goal but totally botches the process
One example that made headlines involved Sebastian Heyneman, founder of a startup in San Francisco. According to a report by the New York Times, he asked an AI agent to land him a speaking opportunity at the World Economic Forum in Davos, Switzerland.
The agent delivered. The spot was secured.
There was just one catch: the cost of participation was around 30 thousand dollars, an amount he could not afford.
The agent completed the mission based on how it interpreted the request, but completely ignored the user’s financial context. In practice, it turned a legitimate goal into an impossible commitment.
The bot Heyneman used was developed by Tasklet, a company that builds agents to automate business tasks. Tasklet’s founder, Andrew Lee, explained that this kind of situation usually comes up when the user gives confusing or conflicting instructions.
According to him, agents can absolutely execute purchasing tasks and do things any consumer would do. But that does not mean this use case is mature enough to be widely adopted.
The perspective from those building these systems
Andrew Lee himself raises an important warning: in his view, using agents specifically for shopping today is still not a great idea.
In an interview with CBS News, he said that agents are, in a word, hard to trust. Even as someone who works directly with this technology, Lee says he does not feel comfortable letting a system control where his money goes on its own. And as a company, Tasklet does not recommend that clients fully delegate purchases to autonomous agents for now.
Another expert interviewed by CBS, Matt Kropp from Boston Consulting Group, echoes that same view. He points out that the use of AI agents for shopping is still not mainstream and, at this stage, is considered high risk precisely because there are not enough safeguards and protections in place yet.
He notes that, technically, an agent could even buy a car if it had access to the right data. But he adds that, in practice, he would not hand his credit card over to the system with that level of freedom.
Security risks: when the agent becomes a target
Beyond interpretation errors and bad financial decisions, there is a very serious layer of security risk involved in using AI agents for shopping.
A critical point is raised by Bretton Auerbach, founder of a tech startup in New York. He draws attention to the ways agents can be tricked by fraudulent websites or malicious content.
Picture this scenario:
- You give the agent your card details,
- tell it to go to a specific site and buy a product,
- and the agent navigates there to complete the transaction.
According to Auerbach, there are ways to fool the agent. For example, it could mistake a legitimate site for a very well-crafted phishing page. If that malicious page displays a flashy message asking it to paste the full card number into a highlighted field, the agent might follow that instruction to the letter without realizing it is being manipulated.
This type of risk is especially serious because the agent does not feel suspicion, does not find a layout strange, and does not question poorly written text. It just follows patterns and instructions. If those patterns are exploited by attackers, the AI can end up becoming a direct bridge between scammers and the user’s financial data.
Why it is still too early to hand everything over to machines
Putting all these points together, the view forming among experts is fairly clear: the technology behind AI shopping agents exists, but the environment around it is not ready yet.
Among the main issues today:
- Lack of guardrails: there are still no standardized, robust mechanisms that safely limit what the agent can and cannot do with your data and your money;
- Ambiguity in instructions: poorly worded requests can lead the agent to make unexpected decisions that are technically correct based on what it understood;
- Vulnerability to scams: agents can be deceived by fake websites, malicious messages, and sophisticated digital manipulation techniques;
- Gray areas of accountability: when something goes wrong, it is still unclear who is responsible in many scenarios — the user, the agent’s company, the card network, or the platform where the purchase was made.
Meanwhile, companies like American Express are starting to position themselves by offering protection against agent errors, which is already an interesting step. But that does not address all the risks in the landscape, especially when it comes to targeted attacks and the misuse of personal data.
So, should I let an AI agent shop for me or not?
AI agent technology applied to consumer spending is promising, with enormous potential to reduce friction, personalize deals, and save time. But looking at what experts directly involved in this ecosystem are saying today, the message is straightforward: it is not yet the right time to hand over full control of your shopping to AI.
Testing assistants that help you research, compare, and track prices can be useful and relatively safe, as long as you keep the final decision in your own hands. But giving an agent full autonomy to spend for you, with open access to your card and little oversight, is something that, for now, a lot of experienced AI professionals still prefer to avoid.
Bottom line: AI agents can already do a lot of things for you, including making purchases. But before diving headfirst into this trend, it is worth taking a careful look at the risks, understanding how these systems work, and most importantly, staying in control of where your money is going and who has access to your data.
