If that question made you pause, you're not alone.
Seventy-two percent of U.S. consumers have used an AI assistant,¹ in addition to search engines to discover products. People are happy to let AI recommend a couch.
But actually letting it hit buy? That’s more complicated.
It’s not a question of AI model capabilities. AI can find products and navigate digital commerce quite well today, and payments are becoming increasingly embedded within the experience.
The challenge is making agentic commerce secure, permissioned, and reliable enough to scale. I believe we can do it — and, when we do, it may represent the biggest shift in digital commerce that I've seen in more than two decades in payments.
The agentic commerce reality has not caught up with the potential. Yet.
A year ago, the vision was compelling: autonomous shopping agents would transform how we all discover, shop and pay.
Over the past year, AI platforms have made real progress, turning product discovery into highly engaging and customized conversations with users. But fully autonomous purchasing? It’s still early days. The vast majority of deployments to date have kept the human-in-the-loop but more fully autonomous pilots are gaining momentum quickly.
This pattern shouldn't surprise anyone. This is how every major commerce shift has unfolded: the technology lands first, and the new behavior follows. We saw it with e-commerce. We saw it with mobile. And we are seeing it again with agentic commerce.
Trust ultimately determines whether a new way to pay scales.
Trust Issues
Today, two trust issues present hurdles to wider adoption. The first is about the trust we can place in the technology underpinning agentic shopping and commerce.
The web was built for people. Merchants have spent years identifying bots, fraud, and suspicious behavior. Now, when an AI agent tries to buy something on behalf of a real consumer, legacy systems often flag it as a threat and not a prospective sale.
Think about what that means. A legitimate agent, acting for a real person, with real money, gets blocked because it looks similar to an attacker.
This poses a whole new set of questions we must address. Is this a good or bad actor? Is it acting for a real person? What has that person authorized? Can the merchant trust the agent? Can the agent trust the merchant? To solve for this trust deficit, we need to deliver transparency of user, agent and seller identity.
The second trust problem is a fundamentally human one. People want to get to know who or what is acting on their behalf before they cede too much control. What did I authorize? What happens if something goes wrong? How do I stay in control? Consumers will need to experience reliable human-in-the-loop agentic shopping and commerce transactions, before leaping into fully autonomous agentic commerce.
Why It’s Still Happening
Despite these hurdles, momentum is accelerating. AI platforms are making it easier to purchase products within their platforms, making the buying process smoother. Merchants are rethinking how agents will interact with their sites, catalogs, and checkout flows. Banks are preparing for new authorization patterns. Developers are building agent-first experiences.
The companies that treat trust as core to the experience will be the ones that win.
Early Signals: B2B May Move First
Some of the first scaled use cases are not consumers asking agents to buy sneakers. They are business use cases: think procurement, supplier onboarding, invoicing, reconciliation, and so on. This work is repetitive, rules-based, and ripe for better automation.
In B2B, the trust model is cleaner: known parties, structured controls, existing commercial rules. That clarity will let business adoption outpace consumer adoption — at least initially.
The same early -adopter logic applies to machine-to-machine payments, developer environments, and embedded commerce flows. Anywhere agents can turn intent into action with less friction.
Where Visa comes in
For more than sixty years, Visa has delivered trust and enabled commerce to scale while complexity and risks continuously increased. We did it when payments moved online. We did it when commerce went mobile. Now we're doing it again as AI agents enter the payment flow.
We’ve partnered with more than 100 companies to integrate AI 'agents' directly into the Visa Intelligent Commerce network. This is more than a pilot program; it’s already powering experiences you can use today, like Amazon’s ‘Buy for Me’ and Meta’s smart autofill. To ensure these transactions are seamless and secure, we’ve deployed new tokenized payment credentials, designed for agentic commerce — essentially digital 'keys' that allow AI to pay without ever exposing your real card details.
And our partners are moving along with us in lock step. More than 150 issuers are currently 'pressure-testing' these payments through our Agentic Ready program. Proving our systems can distinguish between a human and a piece of software, ensuring that when your AI assistant makes a purchase, it’s just as safe as if you had swiped the card yourself.
Closing the trust gap
The industry is collectively working on getting the technology ready. The real question is, will we, consumers, be ready?
Our new Trust Index¹ reveals both a stark reality and an exciting opportunity. While the 'agentic economy' promises total convenience, right now only 23% of U.S. consumers trust GenAI to handle payment transactions on their behalf.
But here’s the kicker: when Visa is the one securing the transaction, that trust jumps to 61%.
Commerce scales when people and businesses trust it. To realize the promise of agentic commerce, we must close the trust gap — that's the work ahead. And that is what Visa does best.
¹Trust Index Methodology
This survey was conducted on behalf of Visa by the Harris Poll using its Omnibus survey platform - a recurring bi-weekly survey tool that can accommodate ad hoc questions at a moment’s notice for quick turnaround.
The survey was fielded May 26 - 28, 2026, in the United States. Respondents are matched to the U.S. General Adult population based on the U.S. Census. The total sample size was 2,065 US consumers. The payment transaction question sample was 1,028 – 1,034 per brand tested.