Are You Ready to Let an AI Agent Shop for You?

By 2029, AI agents will resolve 80 percent of common customer service issues without human intervention, predicts Gartner, a research and consulting firm.

Visa recently announced an Intelligent Commerce platform working with partners such as Anthropic, OpenAI, Perplexity and other generative AI companies. The first agentic transactions under the new platform could happen by the end of summer, Nelsen says.

Rival Mastercard is also working with industry partners on integrating what it calls Agent Pay. Someone planning a milestone birthday party, say, could chat with an agent to curate outfits and accessories based on a consumer’s style preferences, the party venue’s ambience, even the weather forecast. An intelligent agent can take that feedback to make a purchase and suggest the best way to pay.

Avoiding mistakes

To be sure, the technology is still green.

Operator is “able to do a lot of things,” Yash Kumar, who works on new product explorations at OpenAI, told me while I was testing the research preview. But he conceded at the time that “it also fails miserably at a lot of things.”

One early failure was brought to the public’s attention last February by Washington Post tech columnist Geoffrey Fowler after Operator went rogue and spent $31 on a dozen eggs.

OpenAI acknowledges that the risks of the agentic mode remain. Users can access the capabilities from a drop-down menu. You describe what you want the agent to accomplish, and as it begins to perform your tasks, you’ll hear an on-screen narrator let you know what ChatGPT is doing. 

For safety, you can interrupt and take control of the browser at any time, OpenAI said in a blog post. ChatGPT will also ask for your permission before “taking actions with real-world consequences, like making a purchase” or sending an email. ChatGPT is also trained to “actively refuse high-risk tasks such as bank transfers.”

OpenAI said you can delete all browsing data and immediately log out of all active website sessions via privacy controls. It added that it does not collect or store any data you enter during the sessions, including passwords.

A lot less scrolling

In May at its annual I/O conference for industry developers, Google CEO Sundar Pichai announced that the company is starting to bring agentic capabilities to Google Chrome, Google search and the Gemini AI app.

Pichai explained how an emerging tool, Agent Mode for Gemini, could work. He outlined a scenario where roommates are searching for a new apartment in Austin, Texas, each with a limited budget. They want a washer and dryer, or at least a laundromat nearby.

“Normally, you’d have to spend a lot of time scrolling through endless listings,” Pichai said. “Using Agent Mode, the Gemini app goes to work behind the scenes and finds listings from sites like Zillow that match the criteria.”

He added that Gemini can schedule apartment tours, eventually freeing up the roommates to do stuff they’d want to do, “like plan the housewarming party.”

Though you can’t use it to purchase things (at least not yet), you can try another Google AI tool for free: Deep Research can generate lengthy and detailed reports on almost any topic, including subjects of interest to older adults. .

For example, you can ask Deep Research to “Compare and contrast the long-term health benefits of a Mediterranean diet versus a plant-based diet for people over 50, citing recent scientific studies and discussing potential challenges for adoption.”

Or: “Given current market volatility and inflation concerns, what are commonly known retirement investment strategies for individuals approaching or in their early retirement (ages 50-75)?”

You can also generate podcast-style Audio Overviews based on the Deep Research report and hear two AI hosts discuss the subject.

Amazon is testing a “Buy for Me” agentic feature that lets folks using the Amazon Shopping app complete purchases from third-party retailers if Amazon doesn’t directly sell the item. When a consumer taps the Buy for Me button that appears next to certain product search results, Amazon’s AI takes care of the payment and processes the order.

Amazon cautions that its own return policies don’t apply, so you’ll want to make sure you’re comfortable with the brand or site you’re buying from.

Don’t screw up the order

“Everyone will have something at some point in their life to say, ‘Actually, this would be more convenient if you just found this and bought it for me,’ ” says Visa’s Nelsen.

But consumers will have to trust that AI agents won’t screw up shopping orders, will be effective at finding the best values and won’t violate user privacy. Agents will be only as reliable as the vetted or curated data they’ve been trained on.

Shoppers in the Salesforce report emphasized the need for strong data privacy protections, easy on and off switches and the ability to approve purchases before agents go ahead.

Will older adults buy in?

According to the report, older adults are more skittish about working with agentic AI than their younger counterparts are. Gen Z shoppers (born between 1997 and 2012) are by a margin of 63 to 23 percent more likely than boomers (born between 1946 and 1964) to want product recommendations from AI agents.

Visa’s Nelsen says that when consumers tap a credit card to their phone or manually load the information, the 16-digit card number is replaced by a digital token or a number that the company says can be used only by that agent, similar to technology employed in digital wallets like Apple Pay or Google Pay.

The agent logs the specific instructions the shopper provides to prevent it from going off the rails and overpaying for something or buying an unwanted item. Instructions are matched to the purchase the agent is about to make, Nelsen says. ​​

Agents can replace human scammers

One presumed advantage AI agents have over human shoppers: They won’t get bored, says Mark Stockley, a U.K.-based cybersecurity evangelist at ThreatDown by Malwarebytes, which offers security solutions for small and medium-sized businesses.

The dark side: By replacing human scammers, AI agents can automate and accelerate ransomware and other cyberattacks, Malwarebytes warns.

Still, the potential is enormous.

“An agent that’s doing something for you, like checking your blood sugar, [could examine] your shopping list and say, ‘By the way, those three foods you’re picking out — this is the brand you should buy instead, because these don’t have the chemical preservatives,’ ” Palmer says. “That kind of thing, where the guardrails are automated, [could make you] much safer.”