Not too long ago, buying something online meant doing all of the work alone, hunting through site after site until the right product finally turned up. But that routine is changing now that AI assistants have started absorbing work that used to fall entirely to shoppers.
Experts at Shopware have observed fifteen times more traffic arriving at online stores from AI tools during the first quarter of 2026 than during the same stretch of 2025. And Adobe Analytics shows how common that use has become among consumers, finding that 41% of U.S. shoppers used generative AI for online shopping during June 2026.
Letting a tool handle parts of the shopping process on a customer’s behalf is part of what agentic commerce describes, and the spending attached to it is what analysts are now watching. In fact, Morgan Stanley estimates that AI shopping assistants could account for $190 billion to $385 billion of U.S. ecommerce spending by 2030.
But Brian Nowak, who leads the firm's U.S. internet research, said widespread adoption still depends on new products arriving and consumer habits changing. So while shoppers have carried most of the work of buying online since it began, AI agents are now positioned to take a sizable portion of it off their hands. But how far that work moves depends on what these tools can already do.
What are AI shopping agents?
AI shopping agents are software systems built to take a shopping instruction and keep working after the first response. Mastercard describes agentic commerce as allowing an AI agent to "close the loop," or complete a task for someone with limited manual input.
Unlike a basic chatbot, which responds to one request and waits, an agent carries that request through several parts of a purchase. For example, a shopper looking for noise-canceling headphones under $300 states the budget and the feature they care about most. The agent then reviews available choices and returns a recommendation with an explanation.
And parts of that process are running today, with OpenAI and Stripe having developed the Agentic Commerce Protocol to connect AI agents with merchant checkout systems.
Still, direct purchasing through AI assistants remains limited, with OpenAI pulling back from its standalone Instant Checkout experience just five months after it launched. In many current implementations, users still retain final approval before a purchase is completed.
From search engine to shopping agent
The way people shop online looks much different than it did a few years ago, including how they come across brands they have never heard of. Shoppers can now begin with a problem they need solved and a budget, even if they do not know the product name.
INSEAD researchers Nathan Furr and Andrew Shipilov note that people using AI are less likely to search for a specific brand and more likely to ask for the best product for a particular need.
So someone dealing with a leaking kitchen sink, for instance, does not need to know which valve to search for first. Rather, the shopper can begin with the problem itself while the software narrows the choices.
And that behavior is already showing up across survey data, with McKinsey finding that 44% of people who have tried AI-powered search now call it their primary and preferred method, compared with 31% who prefer the traditional way.
Jason Goldberg of Publicis Groupe calls this “the disruption of discovery,” as AI filters which products shoppers see before they start comparing them.
Everyday essentials could be the first major use case
One of the most obvious areas where AI shopping assistants are starting to take on more work is the weekly restock. Groceries and household supplies often come with choices a shopper has already made, and pet food or personal care products follow the same pattern.
Morgan Stanley found that AI-assisted buying is already most common across those categories and expects them to drive the most growth over the next five years.
UVA Darden professor Luca Cian points to what sits underneath those categories, saying AI fits naturally into behaviors people already have and can help "cut through choice overload." For example, a shopper who has bought the same detergent for years has little reason to reconsider dozens of alternatives each time it runs low.
The tendency to stick with a routine also appears in research by Harvard Business School professor Eva Ascarza, who found routine customers remain loyal after a bad experience and are less sensitive to price increases. And those established routines give an authorized AI assistant the information it needs to use past orders and preset preferences to flag when another purchase is due.
AI agents could also change how we make major purchases
A transaction with major financial consequences changes how people shop. Higher prices bring more research, and a refrigerator or a sofa means weighing warranty terms against repair histories across dozens of models.
McKinsey surveyed shoppers across Europe and found 63% already use AI tools to compare brands and prices before buying. And those comparisons stretch well past the price tag once a purchase gets expensive.
In fact, McKinsey lays out a scenario where an agent weighs a $700 furniture shipping bill against selling the item for $200 and replacing it after a move. But comparisons like that get more complicated once long-term costs enter the decision, giving an agent more costs to compare before making a recommendation.
And with thousands of dollars at stake, a useful recommendation has to show how the costs were weighed and why one option came out ahead.
What happens to retailers when AI becomes the customer's gatekeeper?
The reality brands face today is that more people are letting AI assistants narrow their choices before they ever reach a retailer’s website. John Carroll, president of Connected Commerce at Acosta Group, says generative AI tools are becoming “the new gatekeepers of the shopper journey.”
His firm found shoppers may see only two or three options compared with more than 25 on a digital shelf. And with so few products making that cut, a retailer can lose consideration before a shopper ever sees its website. BCG says retailers need product information that is factual and machine-readable so AI systems can evaluate it correctly.
BCG also cites Anne-Claire Baschet, who says agent-ready commerce depends on accurate product data and current stock availability. So retailers now have to make sure software can understand what they sell before traditional marketing gets a chance to work.
The biggest questions. Trust, privacy, and control
Trust is one of the biggest concerns around AI, but handing a software assistant permission to spend puts real money behind those concerns. Visa found nearly nine in ten shoppers want transparency into how an agent makes decisions, while about half would stop using one if they lost control over it. Part of that control is knowing whose interests guide a recommendation.
Consumer Reports argues agents should serve shoppers rather than advertisers and disclose conflicts that could influence what they suggest. And the same demand for control extends to personal data, with Visa finding 85% want control over the data an agent can access.
But deciding what an agent is allowed to know is only part of the problem once it is also allowed to spend. Baker McKenzie says U.S. law still has few cases that explicitly address AI agents, leaving open questions about responsibility when a purchase goes wrong.
Its guidance points toward clear spending limits and human approval before an agent completes purchases that exceed what a shopper has authorized.
The future of shopping may be less about searching and more about delegating
Delegating more of the buying process still has a long way to go. Paul Krauss, Partner AI at Team One, told Shopware that most shopping agents today sit closer to smarter search tools, with real delegation appearing only in isolated cases. But the expectation is that agents take on a larger share of the buying process.
McKinsey senior partner Lareina Yee expects nearly all retailers to eventually serve a significant share of customers through AI agents rather than human users. Even with more agents acting on a shopper's behalf, Krauss says the human still sets the intent while the machine acts on it.
With that intent established, the agent can take on more of the buying work inside the limits a shopper sets. And as more people grow comfortable setting those limits, AI assistants will move further into the buying itself, having already taken over much of the research.
