The AI shopping revolution

First things first: Is this a real, concrete product with a launch timeline, or is it something closer to a concept designed to position the brand in the market narrative?

No, no—it’s real. We have partners actively integrating and giving us constant feedback to make sure we’ll be ready for general availability sometime towards the end of 2025.

What are the main use cases you're imagining?

One simple thing that we’re trying to solve for is a native commerce experience. Any Visa credential, as long as it’s tokenized, should be able to natively work while you’re still in the AI platform that you’re using, like ChatGPT. The idea is the transaction happens on that platform—you’re not sent to another website to make the purchase—and in a few milliseconds, while you’re still in the context window, the AI says, “Here is what was bought, the receipt has been emailed to your inbox, what else would you like me to do?”

How do the use cases increase in complexity from there?

Imagine you’re planning a camping trip. You could say, "My budget is $1,000, I need the best available tent by next Friday"—and as long as it meets these conditions, you don’t care what tent it is. And the AI goes and buys it.

One of the other things we’re super excited about is our personalization capability— we can provide some very insightful and helpful signals for a better discovery experience. If you’re looking for a hotel in Manhattan, and have opted in your card, the AI could analyze the types of hotels you have stayed at and those frequented by others with similar spending patterns, and suggest, "We think you would really love to stay at this place."

Are you imagining the human will be asked for approval before AI makes a purchase?

This is the question we talk about a lot: "What is the intent?" In e-commerce, there’s a clear indication of a consumer intent as you click to say, “I want to buy this. But in a natural language interface, what counts as a real intent to buy? That’s a complicated question. We’re still testing and learning, but we may have it so, at least for the first few transactions, the agent asks the consumer, “Are you sure you want me to buy this?” Now, over a period of time, if you as a consumer don’t want to do that, you may have controls where you can get away from doing it, as long as you have a consistent pattern of how you purchase things.

If an agent makes a high-value purchase that the user never intended, who eats the loss: the user? The merchant? Visa?

You’re asking the all important question. I think that there’s a lot of work for us to do to figure out what that looks like. The way we’re designing this is we’re trying to root out as much fraud, and as much potential of rogue behavior, upfront in the provisioning and the tokenization and authentication of the cards. The scenario where we end up in a debate on liability comes to a very, very, very narrow use case, and then we fall back on our existing world of how e-commerce works. Now, over a period of time, does this evolve where we introduce new vectors of how this works? It’s a very good possibility. But we’re not there yet.

Do you think agents are likely to be paying in fiat currency, or do you think this is going to shift towards stablecoins?

We are big believers in stablecoins. Our view is, regardless of how a consumer chooses to hold money, whether it’s fiat, whether it’s stablecoins, whether it’s a commoditized way of holding money, ultimately there has to be an interoperable layer where the exchange of whatever the form of stored value is needs to happen. Could we see some of these transactions shift to stablecoins because they are a better medium of exchange or value? Probably. And if it is, I think we welcome it, because we are making sure that our network is interoperable, and we are adding a real interoperable value, even with things like stablecoins.