What happened
The RealReal, a company that sells secondhand luxury goods, has updated its anti-bot protection system on its website. After the update, it became considerably harder for artificial intelligence agents to log in to the site, add an item to the cart, and complete a purchase. This change is not a routine protection update, but a response to a specific problem that has emerged around the practice of shopping with agents. The company examined how agents were interfering with the trading process and took steps to limit that interference.
The change was made after a test conducted by The Verge. During the test, Meta's Muse agent acted on a specific task. The agent was instructed to add listings from The RealReal's site to the cart again and again. Each item was held in the cart for only twenty minutes, and as soon as the period expired, the agent immediately added it to the cart once more. In this way, the items remained closed off to other shoppers for several hours, and sometimes even longer.
Why the cart problem is serious
Adding to the cart is usually considered an ordinary step in shopping. But if an agent repeats this step without interruption, the situation changes fundamentally. While an item is in the cart, another shopper cannot buy it. If, once the twenty-minute period expires, the agent adds the item to the cart again, this restriction effectively continues without a break. As a result, real shoppers can see the item, but are left without the opportunity to buy it.
This is a particularly sensitive issue for secondhand trading. If a store has many copies of the same item, one item staying in the cart may not be a big problem. But if an item in the cart stays closed to other buyers for a long time, questions arise about the fairness of the trading process. An agent does not act at human speed, but continuously and without tiring. For this reason, the actions of even a single agent can limit the opportunities of many shoppers.
The Verge test demonstrated precisely this problem. The problem arose not because the agent carried out the task, but because the task itself worked in a way that was not intended for the trading system. The agent does not choose an item like a person, think it over, and then buy it. It automatically repeats the process as soon as the period expires. It is this repetition that disrupts the normal working order of the marketplace.
What the company said
The RealReal's chief technology officer, Luke Friang, said by email that the company had addressed precisely this situation. The case in question concerns the use case involving the Muse agent shown in The Verge article. The company acknowledged the problem and updated its protection system with this specific case in mind.
According to The Verge, The RealReal's chief technology officer, Luke Friang, said by email that the company had addressed precisely the case involving the Muse agent shown in The Verge article.
This statement shows that the company addressed the problem not with general words, but in response to a specific test result. The update is not aimed at completely stopping all actions by agents. It is aimed mainly at strengthening barriers for agents at the stages of logging in, adding to the cart, and payment. In this way, the company is trying to prevent agents from automatically carrying out important stages of the trading process in place of a human.
The company also said it would not stop there. The RealReal stated that, as shopping with artificial intelligence continues to develop, it will keep strengthening its protective measures. This means that the protection system is not a one-time update, but an ongoing process. Agents may try new methods, and the marketplace may introduce new measures against them.
What agents can still do
Importantly, The RealReal has not thrown agents off the site entirely. Agents can still find and display The RealReal listings. They can also direct Muse users to a specific product page. In other words, the search and referral part remains open, but adding to the cart and completing a purchase have become harder.
This distinction matters for the future of agentic shopping. On the one hand, an agent can help a user find a product. On the other hand, the payment and purchase-completion stage appears to need to remain under the marketplace's control. The RealReal has drawn exactly this boundary. An agent can show, but it cannot fully take over the trading process in place of a human.
This approach also shows how the relationship between marketplaces and agent developers may take shape. A marketplace does not reject the agent entirely, but sets the limits of its actions. For the user, this may mean that finding a product becomes easier, but the final purchase decision and process remain under human control.
What shopping means for Meta
At Meta's earnings call for the third quarter of the current month, it was said that shopping has been the most widespread use case for the Muse agent so far. Purchases averaged more than two hundred dollars. It was also reported that within forty days, the app had been downloaded by ten million people in the United States of America.
These figures show that agentic shopping is not a simple experiment. People are actively using the agent to find products and make purchases. The fact that the average purchase amount exceeds two hundred dollars also indicates that serious purchases are being made in this process. And the ten-million-download figure shows that the agent has reached a broad audience in a short time.
Against this background, The RealReal's decision takes on particular significance. At a time when agentic shopping is becoming more widespread, marketplaces are beginning to set their own rules. While Meta tries to increase the agent's convenience, marketplaces are defining who may act on their site, how, and at which stage. Finding a balance between these two interests will remain the central question for the next stage of agentic shopping.
What happens next
Although The RealReal has said it will continue to strengthen its protection, it is still too early to draw firm conclusions about the outcome of this contest. Agent technology is developing, and marketplaces are also updating their protection methods. Each new protection measure may prompt agents to try new ways of acting. And each new agent method may prompt marketplaces to strengthen protection further.
For now, one thing is clear. On The RealReal's site, logging in, adding to the cart, and completing a purchase have become harder for agents. Agents can display listings and direct the user to the product page. The marketplace has said it will continue to strengthen its protection. The future of agentic shopping will be shaped around precisely this boundary. The answers to the questions of who finds the product, who adds it to the cart, and who completes the purchase will depend on an agreement between marketplaces, agent developers, and users.
