This week, I want to talk to you about something that has been on my mind for days. Something that is happening right now that, honestly, sounds like science fiction. But it is not.
Imagine this: you tell your AI assistant, ‘Do your weekly shopping,’ and the agent opens your online supermarket, adds to the cart what you usually buy, applies available coupons, and pays with your card. Without you touching a button. Without you confirming anything.
Visa has just demonstrated that this is already possible. With hundreds of real transactions. With real money. Together with Banco Santander. It’s that simple. And that disturbing…
The week when AI agents started to buy
Less than two months ago, Visa announced that it had successfully completed hundreds of real transactions in which autonomous AI agents purchased products, paid for services, and handled recurring payments. Without a human having to click confirm.
The pilot was carried out in five Latin American countries, with Banco Santander as a partner. And we are not talking about fictitious transactions or sandboxes. We are talking about real money circulating in the real world– not bad for a pilot that, until two weeks ago, most of the world did not know existed.
The moment when everything started to accelerate
Timing matters. Because in the last six weeks, we’ve seen a picture of synergies that I don’t recall ever seeing in the history of payments technology.
Visa not only launched its Intelligent Commerce together with Santander.
Mastercard announced Agent Pay, a system specifically designed for AI agents to execute payments, with confirmed partnerships with Citibank and US Bank.
Stripe created the Machine Payments Protocol, a standard they co-signed with Visa, Mastercard, OpenAI, Anthropic, Shopify and Revolut. Six of the biggest players in the tech payments industry.
And, in addition, Visa has open-sourced its Trusted Agent Protocol on GitHub. So literally any startup can build AI agents that pay for their users.
OpenAI was not far behind. It introduced its own Agentic Commerce Protocol, integrated with Shopify, allowing agents in ChatGPT to shop directly without switching between applications.
In six weeks, the entire payments architecture was reorganized around a single idea: AI agents are going to control trillions of dollars. What if we start preparing for that now?
The problem: trust that simply does not exist
This is where the story gets complicated.
According to Bain & Company, only 24% of consumers would trust letting an AI agent make purchases on their behalf. 24%.
That number has been haunting me for weeks.
And it doesn’t stop there. Twenty-seven percent of consumers outright refuse to share personal data with AI agents. Nearly one in four people have drawn a red line and said, “No trespassing here.”
But here comes the weird part. What really has me confused.
Because 70% of consumers in Latin America already use AI at some point in their buying journey. We are not talking about a rejected technology. We are talking about something that is already mainstream. But, at the same time, it generates a visceral rejection when you mention “an agent that buys for you.”
It’s like he’s saying, “Look, most of you already use ChatGPT every day. But if I tell you that ChatGPT is going to pay your electricity bill… all of a sudden, the trust is gone.”
The dilemma is real.
An Inconvenient Truth: Christmas 2026
Visa has already put a date on the table. December 2026.
The goal: millions of consumers using AI agents to make purchases.
That’s not a “we’d like to.” That’s “let’s do it.”
And the infrastructure is in place. Visa points to AI agents being responsible for a significant volume of transactions this holiday season. Think about it for a second: Black Friday night, you’re asleep, your AI agent is intelligently shopping for gifts because it has learned what your family likes.
Aldar in the UAE, is already using agents for recurring property payments. So it’s not future. It’s present.
The trust vs. inevitability dilemma
The first thing is that there is a colossal clash here between what people say they want and what is bound to happen anyway.
24% trust is a number for today. But three years from now, when 60% of online purchases are made by agents, that number will change because standardization is inevitable. Like autonomous cars, credit cards, and the internet. At first, we say no. At the end, we say we can’t live without it.
The second thing is that Visa, Mastercard, and all these companies are not waiting for trust to exist. They’re building as the trust comes. It’s different. It’s riskier. And that’s probably the right thing to do.
Third, and this is what I can’t get out of my head: what happens when an agent screws up? I’m not talking about a technical glitch. I’m talking about an agent who, through an error of logic, buys 500 cups of coffee when he intended to buy one. Who is responsible? The agent’s creator? Visa? OpenAI? The consumer?
This is not yet clear.
But watch out, it's not all hype
Because there is one detail that Visa is not trumpeting: 27% direct refusals is high. Very high.
When 27% of the population says “don’t touch my payments,” that’s a sign that we’re putting too much speed on an accelerator that maybe we shouldn’t be stepping on.
Regulation, moreover, is blushing as it tries to catch up with us. Europe is looking at this and thinking, “We have to write regulations for something we don’t even know if it’s going to work.” The U.S. is pretty much in “let it self-regulate” mode. Latin America, where Visa did the pilot, has no idea what’s going on.
And frankly, that should concern us. Not paralyze, but worry.
Look, what I think after seeing all this information is that we are at a historic moment of bifurcation. Either we build payment systems with AI agents responsibly, with clear regulation, with limits on what agents can buy without human approval, with full auditability, and with a grievance mechanism when things break….
Or we get caught up in the momentum, launch all this without brakes, and realize in 2027 that we made a pretty profound mistake.
Visa is betting on the former. That’s a good sign. But consumer confidence at 24%… that’s a sign that the communication work hasn’t started yet.
The fact that 70% of Latin Americans already use AI at some point in their purchase process means it is not a rejection of technology. It’s a rejection of loss of control. It’s human instinct to say, “I want to know what’s happening with my money.” And that, frankly, is not unreasonable….
What if that happens in Christmas 2026? What if millions of consumers discover that their AI agent bought something they didn’t want? What if two media cases of fraud or error happen? Does trust survive that?
Or, the other way around, what if it works perfectly? What if the agent is so smart that it understands better than we do what we need? What if by Christmas 2027 it asks “why are you still doing these purchases manually when you could delegate them?”
This is the future that is being built right now…
Have a good week!
SOURCES OF INTEREST
- Visa Intelligent Commerce
- Santander and Visa – First Latin American pilot
- Visa Trusted Agent Protocol – GitHub
- Mastercard Agent Pay + Fiserv – PYMNTS
- Machine Payments Protocol (Stripe + Tempo) – The Defiant
- OpenAI – Buy it in ChatGPT
- Shopify + OpenAI
- Bain & Company – Consumer Trust in AI Commerce
- Visa – Aldar and agentic payments
