The harder part is trusting an AI to actually spend your money.
In a move that quietly reshapes the boundary between human intention and machine action, Visa has embedded its payment infrastructure directly into ChatGPT, allowing AI agents to move from recommending purchases to completing them. Where previous attempts at autonomous AI shopping collapsed under merchant resistance and technical fragility, this partnership divides the labor differently — OpenAI reasons and decides, Visa authorizes and protects. The deeper question being negotiated is not technological but philosophical: at what point does delegating a decision to a machine become indistinguishable from making it yourself?
- The stakes are real — an AI can now reach into your wallet, not just your browser history, marking a threshold that previous chatbot shopping tools never crossed.
- OpenAI's earlier Instant Checkout collapsed in March after merchants rejected its 4% fee and the system proved unreliable, leaving a cautionary wreck on the road this new partnership must navigate.
- Visa is attempting to solve the trust deficit not through the AI itself but through the familiar scaffolding beneath it — fraud monitoring, spending limits, merchant approvals, and dispute resolution.
- Most transactions still require a human tap of approval, but the architecture is explicitly designed to make that step feel unnecessary over time.
- Mastercard is pursuing a narrower business-to-business lane, leaving Visa to make the larger wager on everyday consumer commerce at scale.
Visa has connected its payment network to ChatGPT, turning the chatbot from a shopping advisor into something closer to an autonomous buyer. Tell it you want wireless headphones under $150, and the AI can now find them, select them, and charge your card — without a separate checkout step. That, at least, is the destination.
The road there is littered with earlier failures. Amazon's Alexa was confined to Amazon's own shelves. OpenAI's Instant Checkout let ChatGPT browse and buy across the web, but merchants resisted the 4% fee, errors mounted, and the feature was shut down in March. Visa's approach redraws the map: OpenAI handles the reasoning — understanding what you want and initiating the purchase — while Visa owns everything after, including fraud detection, transaction authorization, and dispute resolution.
Jack Forestell, Visa's chief product and strategy officer, framed the challenge as one of trust rather than technology. People already accept AI recommendations. Trusting it to spend their money is a different matter — and that trust, he argued, must come from the infrastructure underneath, not the AI itself. Visa is calling the system Intelligent Commerce, and it arrives with guardrails: spending caps, merchant approval lists, and required human sign-off on most transactions for now.
But the design anticipates its own evolution. Approve a transaction a thousand times, and eventually the system will ask whether the confirmation step is still necessary. Mastercard is moving in a parallel direction, though more narrowly, focusing on business accounts rather than everyday consumer purchases. Visa is betting on the broader market.
Neither company disclosed fees — a pointed silence given that pricing killed the last attempt. Whether this becomes the future of shopping or another abandoned experiment will depend less on the technology than on who pays, how much, and whether consumers ever fully hand the keys to an AI agent.
Visa has wired its payment network directly into ChatGPT, a move that transforms the chatbot from a shopping advisor into an autonomous buyer. The integration means that when you tell ChatGPT you want wireless headphones under $150, the AI can now find them, select them, and charge your Visa card without asking permission first—or at least, that's the eventual goal.
This represents a significant leap from where AI shopping has been. Amazon's Alexa could only buy things on Amazon. OpenAI tried something similar last year with a feature called Instant Checkout, which let ChatGPT scour the internet for items and complete purchases. But the system was error-prone, and merchants balked at the 4% fee OpenAI charged them. The company shut it down in March. Visa's approach is fundamentally different. Rather than OpenAI handling the payment infrastructure, Visa provides the authorization, fraud detection, and security backbone that allows transactions to happen at scale across any merchant that accepts Visa cards.
The partnership splits the labor cleanly. OpenAI builds the decision-making layer—the technology that lets AI agents understand what you want, find it, and initiate a purchase. Visa handles everything that happens after: confirming the transaction is legitimate, monitoring for fraud, and managing disputes. Jack Forestell, Visa's chief product and strategy officer, framed it as a matter of trust. People are already comfortable with AI recommending products, he said. The harder part is trusting an AI to actually spend your money.
That trust doesn't come from the AI itself, Forestell argued. It comes from the infrastructure underneath—the rules, the security, the process. Visa is building what it calls Intelligent Commerce, a system designed to handle the specific problems that arise when an AI makes a purchase on your behalf. If a customer overspends, buys the wrong item, or later claims they never authorized the transaction, Visa needs a way to sort it out. The company is adding guardrails: spending limits, required approval steps, and lists of approved merchants. For now, most transactions will still require human sign-off. A notification pops up, you approve it, the purchase goes through. But Forestell acknowledged the eventual trajectory. Do that a thousand times, and at some point the AI will ask: do you really need me to keep asking?
Mastercard, Visa's largest competitor, is moving in a similar direction but on a smaller scale. It's focused on business-to-business transactions—allowing a coffee shop to authorize an AI agent to buy advertising services, for instance. Visa's bet is broader: consumer shopping, at any merchant, for anything.
Neither company disclosed the financial terms of their partnerships or what fees customers and merchants would face. That silence is notable, given that Instant Checkout's 4% merchant fee was a primary reason for its failure. The question of who pays and how much will likely determine whether this actually becomes the future of shopping or another abandoned experiment in AI commerce. For now, Visa and OpenAI are betting that people will grow comfortable enough with AI agents handling their money that the infrastructure question becomes secondary. The real work is building the trust first.
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, Visa chief product and strategy officer
Making the leap from having AI agents recommend what to buy to doing the purchasing just requires a whole different level of trust.— Jack Forestell, Visa
A Conversa do Hearth Outra perspectiva sobre a história
Why does Visa need to be the one handling the payment part? Couldn't OpenAI just do this themselves?
OpenAI tried. Instant Checkout failed partly because merchants saw the 4% fee as too expensive, but also because OpenAI didn't have the fraud detection and dispute resolution infrastructure that a payment network like Visa has built over decades. When an AI buys something and the customer later says it wasn't authorized, someone has to decide who's right. Visa knows how to do that at scale.
So the real barrier to AI shopping isn't the AI itself—it's the payment plumbing?
Exactly. The AI part is almost solved. ChatGPT can already find products and understand what you want. The hard part is making sure the transaction is secure, that fraud gets caught, and that disputes get resolved fairly. That's what Visa brings.
The article mentions guardrails like spending limits. How does that actually work? Does the user set them once and forget about it?
Probably. You'd link your Visa card to ChatGPT, set a limit—maybe $100 per transaction or $500 per month—and then the AI operates within those boundaries. At first, it still asks for approval. But over time, as you approve the same kinds of purchases repeatedly, the system learns your patterns and eventually might stop asking.
That sounds like it could go wrong pretty quickly. What if the AI misunderstands what you want?
That's the real risk. Visa is betting that their fraud monitoring catches the mistakes before they become problems, and that their dispute process is fair enough that customers trust it. But you're right—there's a gap between what the company hopes will happen and what actually will when millions of people start using this.