Using an AI agent to shop for you is a strange experience. Tech companies have marketed them as personal assistants: the democratization of work that would normally only be available to the rich and powerful. In practice, at least for buying things, it’s more like sending an anxious and apologetic teenager to the store: they require an excessive amount of instructions and specifications, encounter technical problems regularly, and return with solutions that aren’t exactly what you asked for. Technically speaking, they completed a task, but not the specific one you asked of them and often not in the way you would have approached it.
Meta, Google, OpenAI, and others have introduced some version of agent shopping tools that promise to make buying things even easier than it already is. Demonstrations show AI agents who can seamlessly book flights and hotels, scan billions of products to find the right one, and secure tickets to sporting events on behalf of a human being. But, as AI companies (and their users) are discovering, buying things requires the other party to be willing to sell them in the first place.
The most prominent example of a retailer or service provider refusing to let an AI agent in has been Amazon. The retail giant sued Perplexity, and Amazon said the artificial intelligence company had ignored its requests to prevent its agent from shopping on the site. More recently, Amazon prevented Meta’s Muse agent from purchasing from Amazon. E-commerce sites aren’t the only platforms blocking agents: JPMorgan Chase CEO Jamie Dimon said the bank wouldn’t allow outside agents to access accounts, saying it was “too risky” for customers and the company, and some users report they haven’t been able to use agents to book travel from providers like Delta and Amtrak. AI agents can also get caught in the web of general bot blocking systems, meaning that even if a site doesn’t have a blanket ban, existing firewalls can make agents inconsistent and, by extension, less useful for the types of tasks that tech companies tout as the future.
Even without hard blocks, agent access to retailers is spotty. For example, Muse, who I asked to find a specific facial moisturizer, got caught in Target’s robot check one day, but was able to beat it the next day. When I looked for the best price on granola bars, it suggested a listing from a grocery chain that doesn’t ship to my address. When I asked for mattress recommendations for a certain price, Target blocked Muse again, even though I was able to exit the site a few minutes earlier.
An agent’s success in accessing a site may depend on how they performed the task and what safeguards are in place, Target spokeswoman Loni Monroe said. The edge in an email. (Target has AI purchasing partnerships with OpenAI, Google, and Microsoft.) Agents can access one part of the site but not another, as a way to protect customer data and the shopping experience. Additionally, Monroe said Target has security measures in place that prevent unauthorized agents and robots from accessing customer accounts and initiating transactions on behalf of a human.
It’s also impossible to ignore Amazon’s lack of options within agents. Sometimes agents will say the same thing: When I asked Muse to find the best granola bar price by the piece, it offered options from Sam’s Club, Walmart, and Costco, but acknowledged one big caveat: Amazon might have had the lowest price option, but it was blocked from browsing, so Muse couldn’t confirm. Between waiting for a response, telling Muse to look for more sites, and clarifying that a West Coast-based retailer wouldn’t ship to New York, it probably would have been quicker to search myself. It also explains why AI companies would very much like to have Amazon on board, and why Amazon so fiercely protects how customers shop on its platform. It’s not that Amazon is against robots shopping instead of humans: the company has its own agent shopping feature on the platform that monitors prices, automatically reorders items, and purchases products based on rules set by human users. Another feature that uses an agent to buy products from other sites without leaving Amazon has angered smaller businesses, who say they did not consent to have their storefronts removed and featured on Amazon.
Amazon spokesperson Maxine Tagay confirmed that the company “requested that Meta remove Amazon” from Muse.
“We think it’s pretty straightforward that third-party apps that offer to make purchases on behalf of other companies’ customers should operate openly and respect service providers’ decisions about whether to participate,” Tagay said in an email.
I asked Muse to bulk add the 10 newest listings from a favorite brand to my cart.
It’s one thing for agents to offer comically spotty and disjointed services, but another is a potentially more frustrating experience when they actually work, because of the implications for humans. To test a shopping task I’d really like to automate, I unleashed Muse on The RealReal, a luxury secondhand fashion retailer. As secondhand shopping has grown in popularity, shopping on The RealReal has become increasingly competitive, to the point that the company sells a $12-a-month subscription that allows shoppers to get early access to new items. At The RealReal, part of the checkout process is zero-sum: Placing an item in your cart reserves it for 20 minutes, making it impossible for other people to move past it. I asked my Muse agent to bulk add the 10 newest listings for a favorite brand to my cart, effectively blocking other shoppers from purchasing them before I had a chance to review the latest selections. I had no intention of purchasing any of these items, but I asked Muse to add the items back to my shopping bag each time the 20-minute wait expired. This went on and on all day (I finally finished the task).
If I were a less honest and pure-hearted person, I would send an agent to manage this personalized and technically enabled reservation system endlessly. But as I watched items continually return to my cart, exasperation flooded my psyche, on behalf of the other buyers interested in these items, but also for my future self, who could soon be competing with someone else’s agent doing the same thing. Maybe I already am.
RealReal sees AI agents as an opportunity and is investing in its own AI-powered services while protecting the integrity of the platform, said chief technology officer Luke Friang. The edge in an email.
“Our approach is to remain open to third-party AI agents, allowing them to browse, log in and interact with our marketplace, while maintaining safeguards against automated activities that could disrupt the customer experience,” Friang said.
Sure, regular online shopping can be tedious too – the endless variety of options on the web can make finding the right thing like searching for a needle in a haystack. Agent shopping is the opposite: it’s like looking at the world through a pinhole. If I don’t like the five options Muse has selected for me, I have to ask it to find five more, and so on. I need to give the agent specific parameters and rules, and expand and modify them as needed, little by little. A more routine task, like finding the lowest price on a product you already know you want, is where an agent could be really helpful. However, without Amazon, even that seems as if the task was carried out with incomplete information.
If you have preferences about literally anything you spend money on (price, return policy, shipping rates, a company’s ethical standards, fabric composition), an agent is simply another entity to express those preferences to. With time and the patience to hold my hand, Muse could help deepen the search for something specific: for example, I uploaded photos to Muse of a vintage dress I was looking for, along with the specific year it came from, a list of keywords associated with it, notes about similar items, and second-hand platforms to start searches. Muse couldn’t find the dress for sale anywhere, but alerted me to an upcoming auction of items from the same brand.
The joy and excitement of discovery has largely disappeared.
In an email, Meta spokesperson Tracy Clayton emphasized the partnerships the company has forged with retailers and the control that human users have over the agent’s actions.
“Importantly, users can take control at any time, Muse is designed to always ask permission before purchasing and applies ethical browsing principles when asked to do things a human couldn’t do, such as buying all the tickets to an event,” Clayton said. “The response from retailers has been overwhelmingly positive and we look forward to working with them as we continue to grow Muse and improve the shopping experience.”
There’s another thing that bothers me no matter how long I tinker with shopping agents: the joy and excitement of discovery has largely disappeared. Buying things is an American pastime. We do it in person, online, with strangers on Facebook, and with professional sellers on live streams. We buy things we need and we buy things we don’t, things we know we want and things we have never seen before and that have suddenly become very important. It’s no exaggeration to say that for many, shopping is a pastime, one that has become firmly intertwined with another, more modern pastime, that of scrolling on a phone. We not only covet material belongings; we imbue them with markers of identity and social status. Buying things is intensive: in resources, in work, in emotion. Artificial intelligence companies see this hobby as something that can be optimized. Many people would probably prefer it not to be that way.
