AI agents become active participants in the economy
In a move that edges artificial intelligence closer to full economic agency, Visa has embedded its payment infrastructure into ChatGPT, allowing the system to complete purchases on behalf of users across millions of merchants worldwide. This crosses a quiet but consequential threshold: the AI is no longer merely a counselor but a participant, capable of spending real money in the real world. The partnership arrives as a corrective to OpenAI's earlier failed commerce experiment, this time distributing responsibility between two institutions whose reputations depend on getting it right. Whether trust in the system will prove warranted remains the open question at the heart of this arrangement.
- An AI agent can now find a product, authorize a transaction, and charge your Visa card without you ever touching a checkout button.
- OpenAI's previous attempt at AI shopping collapsed under merchant resistance and a 4% fee structure, making this second effort carry the weight of a necessary redemption.
- The integration theoretically reaches every Visa-accepting merchant on earth — a scale that dwarfs anything Amazon's Alexa or OpenAI's Instant Checkout ever attempted.
- Spending limits, merchant approval controls, and fraud monitoring exist as guardrails, but none have been stress-tested against the real chaos of consumer commerce.
- Fee structures for both customers and merchants remain undisclosed, and the liability question — who pays when the AI buys the wrong thing — is still unanswered.
Visa has connected its payment network directly to ChatGPT, enabling the AI to shop and complete purchases on a user's behalf — find the wireless headphones, charge the card, done. It is a meaningful shift: the chatbot moves from advisor to actor in the economy.
OpenAI had tried this before. Its Instant Checkout feature launched late last year as a digital personal shopper, but merchants rejected the 4% transaction fee and the system proved unreliable. It was quietly shut down by March. This time, the architecture is different. OpenAI handles the intelligence — understanding intent, making decisions, triggering purchases. Visa handles the money infrastructure, fraud detection, and authorization. The division of labor is also a division of accountability.
The reach is far broader than any prior attempt. Where Alexa was confined to Amazon and Instant Checkout worked with only a few enrolled merchants, this integration opens the door to any of the millions of retailers worldwide that accept Visa. Neither company has disclosed what fees will apply — an omission that echoes the opacity that doomed the last experiment.
Guardrails exist: spending limits, pre-approved merchant lists, required authorization steps for certain transactions. But they are untested. The risks are real — wrong purchases, unauthorized charges, fraud disputes — and the question of who absorbs the cost when something goes wrong remains open. Mastercard is pursuing similar territory, though more narrowly focused on business-to-business transactions, sidestepping the thornier consumer protection questions Visa is now confronting head-on.
What has been built is a working prototype of a future where AI doesn't just recommend — it acts, with your money, in the world. Whether that future feels like convenience or exposure will depend on whether the guardrails hold, and whether the financial terms prove fair to everyone the system touches.
Visa has woven its payment network directly into ChatGPT, crossing a threshold that transforms the chatbot from a tool you consult into an agent that can shop and spend money on your behalf. The integration, announced Wednesday, means you can tell the AI what you're looking for—say, wireless headphones under $150—and it will find a pair, complete the purchase, and charge your linked Visa card without you lifting a finger to check out.
This is not the first time OpenAI has tried to turn ChatGPT into a shopping assistant. Late last year, the company launched Instant Checkout, a feature that let the chatbot hunt across the internet for specific items like a digital personal shopper. The experiment failed. Merchants balked at the 4 percent transaction fee OpenAI imposed, and the system itself was error-prone. By March, OpenAI had quietly retired it.
Visa's approach is fundamentally different. Rather than OpenAI building and monetizing the commerce layer itself, the two companies are splitting the work. OpenAI provides the artificial intelligence that lets agents understand what you want, make decisions, and trigger purchases. Visa supplies the payment infrastructure, fraud detection, and authorization systems needed to move real money at scale. Jack Forestell, Visa's chief product and strategy officer, framed it as a matter of trust: "As AI agents become active participants in the economy, Visa's focus is to ensure transactions are trusted, secure and seamless."
The scope is dramatically wider than anything attempted before. Amazon's Alexa could only buy things on Amazon. OpenAI's Instant Checkout worked with a handful of enrolled merchants. This integration theoretically opens the door to any retailer that accepts Visa—potentially millions of merchants worldwide. Neither company has disclosed what fees customers or merchants will pay, a silence that mirrors the opacity that sank Instant Checkout.
The guardrails are real but untested. Visa says the system will enforce spending limits, require approval steps for certain transactions, and restrict purchases to pre-approved merchants. These constraints exist because the risks are substantial. A customer could wake up to find the AI bought the wrong item, or spent far more than intended, or made a purchase the customer never authorized. Banks worry about fraud claims when an AI agent uses someone's credit or debit card. Retailers worry about chargebacks and disputes. The question of who bears the cost when something goes wrong—the customer, the merchant, or the payment network—remains unanswered.
Visa is not alone in this space. Mastercard has been building its own AI shopping features, though on a smaller scale. Mastercard's focus is on business-to-business transactions: an AI agent authorized by a coffee shop, for instance, could autonomously purchase advertising services from web providers to launch a campaign. That's a narrower use case, but it sidesteps some of the consumer protection landmines that Visa is now navigating.
What Visa and OpenAI have built is a proof of concept for a future where artificial intelligence doesn't just advise you—it acts for you, spending your money in the real economy. Whether that future is convenient or unsettling depends largely on whether the guardrails hold, and whether the financial terms prove acceptable to the merchants and customers who will bear the real costs of the experiment.
Citações Notáveis
As AI agents become active participants in the economy, Visa's focus is to ensure transactions are trusted, secure and seamless.— Jack Forestell, chief product and strategy officer at Visa
A Conversa do Hearth Outra perspectiva sobre a história
So the AI can just spend your money now? How is that not terrifying?
It's designed with limits—spending caps, merchant whitelists, approval steps for bigger purchases. But you're right to be unsettled. The real question is whether those guardrails actually work in practice, and what happens when they fail.
Why did OpenAI's version fail so badly?
Two things. They charged merchants 4 percent per transaction, which was too expensive. And the system made mistakes—it would find the wrong product or misunderstand what you wanted. Merchants didn't trust it.
So Visa is betting they can do it better?
Visa has the fraud detection and payment infrastructure already built. They're not trying to invent payments from scratch. But they still haven't said what they'll charge merchants, which is the same problem that killed Instant Checkout.
What happens if the AI buys something I didn't authorize?
That's the unsolved problem. Visa says there are guardrails, but the liability question—who pays if something goes wrong—is still unclear. Banks are nervous about fraud claims. Customers are nervous about overspending.
Is anyone else doing this?
Mastercard is, but they're focused on business purchases. A coffee shop could authorize an AI to buy advertising services. That's simpler because it's B2B, not consumer-facing.
So this is really about whether AI agents become economic actors?
Exactly. Right now, AI recommends things. This lets it actually transact. That's a different world.