As AI agents become active participants in the economy
In San Francisco this week, Visa and OpenAI formalized a partnership that allows ChatGPT to complete purchases autonomously on behalf of users across any Visa-accepting merchant worldwide — a quiet but consequential moment in the long story of human beings delegating economic agency to tools they have built. Where previous attempts at AI-powered commerce faltered on cost and trust, this arrangement divides the labor: OpenAI reasons and decides, Visa authenticates and protects. The deeper question it raises is not whether the technology works, but whether the act of choosing — even something as small as a pair of headphones — is something we are ready to surrender.
- Visa has embedded its payment infrastructure directly into ChatGPT, enabling the AI to shop and pay across millions of merchants without requiring user confirmation for each transaction.
- OpenAI's previous attempt at autonomous checkout collapsed under merchant resistance and a 4% transaction fee, making Visa's established payment relationships the critical missing piece.
- Banks and retailers are bracing for a surge in disputes — customers blaming AI agents for unauthorized purchases, and merchants absorbing the cost of machine-initiated returns and chargebacks.
- Visa has introduced spending limits, merchant approval lists, and tiered authorization steps as guardrails, though the fee structure for merchants and consumers has not yet been disclosed.
- Mastercard is moving in the same direction but targeting business-to-business transactions, signaling that the entire payments industry views autonomous AI commerce as an approaching inevitability.
Visa has connected its payment network directly to ChatGPT, allowing the AI to shop and complete purchases on a user's behalf across any merchant that accepts Visa — a significant expansion beyond the single-retailer limitations that have defined earlier experiments in autonomous commerce. The mechanics are straightforward: a user links their card, describes what they want, and the AI finds, compares, and buys without requiring clicks, form entries, or per-transaction approval. OpenAI handles the reasoning and decision-making; Visa supplies the fraud detection and authorization infrastructure that makes it trustworthy at scale.
This is not OpenAI's first attempt at the idea. Its Instant Checkout feature, launched last year, promised a similar experience but stumbled badly — merchants resisted a 4% fee on every transaction, errors were common, and the feature was quietly retired by March. Visa's involvement sidesteps the fee problem by routing payments through its own existing merchant relationships, a domain where it carries decades of credibility.
Jack Forestell, Visa's chief product and strategy officer, described the partnership at a San Francisco event as a natural evolution: as AI becomes an economic actor rather than merely a tool, payment networks must ensure those actions are secure and seamless. But the risks are not abstract. Banks worry about fraud claims from customers who blame their AI for unwanted charges. Retailers worry about machine-initiated returns and chargeback costs. Visa says it has built in protections — spending caps, merchant approval lists, and stepped authorization for larger purchases — though it has not disclosed what it will charge for the service.
Mastercard is pursuing a parallel path, focused for now on business use cases rather than consumer shopping. Together, the moves suggest the payments industry has already decided autonomous AI commerce is coming. What remains genuinely open is whether consumers will embrace handing their wallets to an algorithm, or whether the small friction of human confirmation turns out to be something they did not want to lose.
Visa has woven its payment network directly into ChatGPT, creating a system where artificial intelligence can shop and pay on your behalf without asking permission for each individual purchase. The move represents a significant leap forward in autonomous commerce—no longer confined to a single retailer or a handful of enrolled merchants, but potentially spanning any business that accepts Visa cards worldwide.
The partnership works like this: you link your Visa card to ChatGPT, then tell the AI what you're looking for. You want wireless headphones under $150? The chatbot finds them, compares options, and buys a pair without requiring you to click through to a website, enter payment details, or confirm the transaction. OpenAI provides the artificial intelligence that makes decisions and initiates the purchase. Visa supplies the payment authorization and fraud detection infrastructure needed to handle this at scale—the trust layer that makes it work across millions of transactions.
This is not OpenAI's first venture into AI-powered shopping. Last year the company launched Instant Checkout, a feature that let ChatGPT scour the internet like a personal shopper and complete purchases on your behalf. It sounded promising in theory. In practice, it stumbled. The system was error-prone, and merchants balked at the cost: OpenAI charged them 4 percent of every transaction's value. By March of this year, the company had quietly retired the feature, unable to gain traction. Visa's approach sidesteps that problem by handling the payment processing itself, a domain where it has decades of expertise and existing relationships with retailers.
Jack Forestell, Visa's chief product and strategy officer, framed the partnership as a natural evolution. As artificial intelligence becomes an active economic participant—not just a tool you use, but an agent that acts on your behalf—payment networks need to ensure those transactions are secure, trustworthy, and seamless. He spoke at a company event in San Francisco on Wednesday, describing a future where AI agents handle routine shopping tasks the way they might handle email or calendar management.
But the risks are real, and both banks and retailers are watching carefully. What happens if the AI agent makes a mistake and buys the wrong item? What if you overspend because you set the guardrails too loose? What if you dispute a transaction, claiming you never authorized it? Banks worry about fraud claims multiplying as customers blame their AI agents for unauthorized charges. Retailers worry about chargebacks and the cost of processing returns initiated by machines rather than humans.
Visa says it has built in protections: spending limits you can set, approval steps for larger purchases, and the ability to designate which merchants the AI can shop from. These guardrails are meant to give you control while still allowing the AI to operate autonomously within boundaries you define. The company did not disclose how much it will charge merchants or consumers for the service, a detail that matters enormously given what happened with Instant Checkout.
Mastercard, Visa's largest competitor, is pursuing a similar path but on a smaller scale. It has announced that AI agents will be able to make purchases on behalf of businesses—a coffee shop, for instance, could authorize an AI agent to buy advertising services and web design from multiple vendors to launch a campaign. That's a narrower use case than what Visa and OpenAI are attempting, but it signals that the entire payments industry sees autonomous AI shopping as inevitable.
What remains unclear is whether consumers actually want this. The convenience is obvious. The risks are equally obvious. The next few months will show whether people are willing to hand their wallets to an AI, or whether the friction of authorization and approval is a feature, not a bug.
Citas Notables
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
La Conversación del Hearth Otra perspectiva de la historia
Why does Visa care about being embedded in ChatGPT specifically? Couldn't any AI do this?
Visa cares because ChatGPT is where the users are. It's the dominant AI interface right now. But you're right—the technology itself isn't unique to OpenAI. What matters is that Visa gets to be the payment layer for the most popular AI agent on the planet.
So what actually went wrong with Instant Checkout? It sounds like it should have worked.
It did two things wrong at once. The system itself was unreliable—it made mistakes, recommended the wrong products, got confused. But more importantly, OpenAI charged merchants 4 percent per transaction. That's expensive. Merchants decided the hassle wasn't worth the cost.
And Visa thinks it can do better because it already has relationships with merchants?
Exactly. Visa doesn't need to convince merchants to join. They're already in the Visa network. Visa just needs to make the payment part work smoothly and keep fraud down. That's what they know how to do.
What's the actual risk here for a consumer? I set a spending limit and approve certain merchants. What goes wrong?
The AI could still make mistakes within those boundaries. Buy you the wrong headphones. Misunderstand what you asked for. And then you're in a dispute with the merchant about whether you really authorized it. Banks are nervous because they'll be the ones handling those disputes.
Is this inevitable? Will we all be letting AI shop for us in five years?
That depends on whether it actually works better than just shopping yourself. Convenience matters, but so does control. If the AI makes mistakes or costs you money, people will turn it off. The guardrails Visa is building in—those are admissions that they're not sure either.