Artificial intelligence is starting to change that pattern. AI “agents,” tools that can act on a person’s behalf rather than simply answer questions, are increasingly being used to plan meals, keep track of household supplies, suggest how much to buy, and automatically restock essentials. The goal is a shopping trip with less guesswork, less waste, and a closer match to each household’s budget and needs.
From comfort to delegation
Consumers are growing more at ease with the technology. According to Accenture’s latest Consumer Pulse Research, 57% of consumers now use AI every week. In grocery, 68% say they are comfortable letting AI agents handle shopping-related tasks, and nearly a third would allow an agent to make the final purchase decision before they pay. A smaller group, 9.5%, say they are ready to hand over their grocery shopping entirely.
Many also want agents to shop around for them, with 61% of shoppers saying they would like an AI agent to compare multiple grocery stores, for example, buying produce at one retailer, snacks at another, and cleaning products at a third, based on real-time prices and availability.
Major companies are already moving in this direction. Amazon’s AI assistant, Rufus, can monitor product prices, notify customers when an item drops to a set price, and complete the purchase. Walmart has launched a conversational agent called Sparky that helps customers find products, read reviews, and place orders. American Express recently announced protections for cardholders who shop through certain AI agents, including verifying an agent’s identity at checkout and covering eligible customers for charges caused by agent errors. Data from Statista shows that about a quarter of Americans ages 18 to 39 have already used AI to research products or shop.
Buying better, not just buying more
For many shoppers, the appeal goes beyond convenience. For 45% of grocery consumers, budget and value are their top priorities, and AI agents can do more than hunt for the lowest price. By taking into account household size, upcoming schedules, dietary needs, and what is already at home, they can help people buy with much greater accuracy.
That precision is already showing results. Among active generative AI users, 26% say AI has led them to choose a cheaper grocery item, while 21% say it has increased the overall size of their basket. Agents can also suggest recipes built around ingredients already in the kitchen, recommend the right package sizes, and hold off on reordering items that are still in stock. Together, these features can cut down on food waste and help shoppers feel more confident about what they are spending.
When things go wrong
Despite the promise, technology experts warn that handing important decisions to AI carries real risks. Agents can misunderstand instructions, cost people money, and potentially give hackers access to personal data. These concerns are especially serious when agents are allowed to make purchases.
Matt Kropp, an AI expert at Boston Consulting Group, told CBS News that agentic shopping is not yet mainstream and remains risky because the system lacks enough safeguards. He said he would not simply hand an agent his credit card.
One widely reported mishap, reported in a New York times article, shows what can happen. Sebastian Heyneman, founder of a San Francisco tech startup, asked an AI agent to get him a speaking slot at the World Economic Forum in Davos, Switzerland. The agent succeeded, but the spot came with a $30,000 fee that he could not afford. Andrew Lee, founder of Tasklet, the company behind the agent Heyneman used, told CBS News that these problems can occur when a user’s prompt contains conflicting instructions.
Lee said today’s agents are fully capable of shopping for people, but he advised against it for now, explaining that the agents are still hard to trust and that his company does not recommend using them to spend money.
Security is another major concern. Bretton Auerbach, founder of LocalMovers.com, told CBS News that bad actors can trick AI agents into handing over a consumer’s personal information. For instance, an agent sent to buy something online could mistake a phishing site for a legitimate one and enter the user’s credit card number when prompted.
Big banks raise red flags
Some of the world’s largest banks share these worries. In a report titled “Building Trust in Agentic Commerce,” six major global banks, including Bank of America and Capital One, warned that agentic shopping carries risks related to transparency, safety, privacy and data, choice, and interoperability, and that those risks grow as agents are given more independence.
The banks noted that consumers are unsure whether AI agents will act in their best interest, fearing that agents could buy the wrong item, overspend, or lose their money to scams. The report also warned that agents might favor certain products or payment methods because of incentives such as higher commissions or lower operating costs, rather than what is best for the customer. It pointed to the potential for higher rates of scams, fraud, and payment disputes, and cautioned that any data breach could expose especially sensitive information belonging to both shoppers and merchants.
Recent events highlight these cybersecurity threats. Meta announced it had found and fixed a zero-day vulnerability in Muse, its new AI assistant that promotes shopping among its features. The flaw could have allowed an attacker to take control of the assistant and exploit any permissions the user had granted. Just a day earlier, Amazon had asked Muse to stop using its site, saying the agent failed to identify itself while shopping and raising concerns about how customer credentials and account data were being handled.
Still, the banks are not rejecting the idea. They said agentic commerce could eventually become a mainstream way for consumers and merchants to do business, and that they are excited about its potential. First, however, the industry needs standards, policies, and consumer protections that have so far lagged behind the technology. The group is now preparing a follow-up paper on how to put its five principles into practice.
