# AI Agents Can Now Spend Your Money. The Hard Part Is Getting You to Let Them > The payment rails that let software agents buy on their own are already live at Visa, Mastercard and beyond. The missing piece is trust, and the rules that would make handing over your wallet feel safe. - Source: Standard - Canonical URL: https://standard.de.com/article/ai-agents-can-now-spend-your-money-the-hard-part-is-getting-you-to-let-them - Author: Standard Staff - Section: Innovation - Published: 2026-07-18T12:00:00.000Z - Updated: 2026-07-18T12:00:00.000Z - Tags: Agentic Payments, AI Agents, Visa, Mastercard, Fintech, Stablecoins --- The technology that lets an artificial intelligence agent reach into your account and pay for something is no longer theoretical. Over a few weeks this summer the biggest names in payments switched it on. What has not caught up is the harder currency the whole idea runs on, which is trust. The plumbing is ready. The question is whether people, and the regulators watching over them, will actually let the machines press the buy button. The scale of the prize explains the sudden rush. Morgan Stanley has estimated that shoppers who are really software agents could account for somewhere between 190 billion and 385 billion dollars of United States online commerce by 2030. That is a market worth building rails for, and the incumbents have decided they would rather lay the track than be bypassed by it. ## The rails went live almost overnight In June, Visa said it would wire its payment network directly into ChatGPT through a partnership with OpenAI. The design leans on tokenization and a set of permissions the user defines up front, including spending caps and thresholds above which the agent has to stop and ask for a human yes. On the very same day, Mastercard unveiled a service built for machines paying machines, capable of settling amounts as small as fractions of a cent, recording the permissions on public blockchains such as Polygon, Solana and Base, and settling across cards, bank accounts and regulated stablecoins. More than 30 companies had signed on by launch. The frontier runs further still. Late in June the crypto exchange OKX opened a marketplace where agents can hire and pay one another with no person in the loop at all. Agents earn portable on chain reputations, settle in stablecoins, and can hold funds in escrow when a job is complicated enough to warrant it. The use cases sound mundane precisely because they are so plausible, an agent paying another to clean a dataset, draft a market report, spin up ad variations, or watch a wallet for suspicious activity. ## Where the trust breaks down The gap between a recommendation and a transaction is where the whole thing gets uncomfortable. People have grown used to AI suggesting a flight or a pair of shoes. Letting it book the flight, spend 2,000 dollars on a holiday, and put the charge through without a final look is a different act of faith. The failure modes are easy to picture, a wrong booking, a subscription that renews without permission, a payment routed to a fraudulent seller, a counterfeit good that arrives instead of the real one. When money moves and something goes wrong, the immediate question is who is on the hook. Few people have watched consumer money habits more closely than Marianne Lake, the former head of consumer and community banking at JPMorgan Chase. Speaking to investors, she offered a blunt reality check, saying she does not think people are going to hand their purchasing over to agents just yet. Her caution was not about the wiring but about the human conditions around it, the unease that surfaces the moment money is involved. ## What would have to be true first Lake laid out the conditions that would have to hold before mainstream shoppers relax. Consequential decisions still need a human in the loop. Agents need to be transparent about what they did, with an audit trail a person can follow after the fact. There has to be a clear framework for liability when an agent makes a mistake, and a dependable way to dispute a charge and get money back. Underneath all of it sits the plain matter of whether customers are actually ready, a readiness the industry cannot manufacture by press release. None of this is a verdict against agentic payments. It is a description of the bridge still to be built. The engineering side of the problem has been answered with striking speed, with tokenized permissions, spending limits and blockchain settlement arriving faster than most expected. The remaining work is quieter and slower, the work of convincing a cautious public and a watchful set of regulators that an agent spending on your behalf is safe, accountable and reversible. Whoever wins the next phase of commerce will be the one that closes that trust gap, not merely the one that opened the account. --- Originally published by Standard. Free to cite with attribution and a link to https://standard.de.com/article/ai-agents-can-now-spend-your-money-the-hard-part-is-getting-you-to-let-them.