AI agents are no longer just answering questions. They are booking flights, comparing prices, and completing purchases without a human clicking "buy." McKinsey estimates this kind of automated shopping, known as agentic commerce, could handle between 3 trillion and 5 trillion dollars in transactions worldwide by 2030.
The problem is not whether AI agents can do this. They already can. The problem is that nobody has a reliable way to check who is really behind the purchase. When your AI assistant buys a hotel room on your behalf, the bank, the merchant, and the payment network have no clean way to confirm you actually approved it, or that a scammer did not hijack your assistant to spend your money instead.
This is not hypothetical. Fraud teams are already tracking fake AI shopping agents and counterfeit merchant sites built specifically to trick real shopping agents into buying from them. Visa has reported that AI driven traffic to US retail websites jumped by 4,700 percent, a sign of how fast this shift is moving, and how far ahead of the rulebook it already is.
Every major payment company wants to write that rulebook itself. Mastercard has its own Agent Pay system and a framework called Verifiable Intent. Visa built something called Trusted Agent Protocol. Google has its own Agent Payments Protocol. OpenAI and Stripe built a separate one called the Agentic Commerce Protocol, which already powers real purchases through ChatGPT from Etsy sellers, with Shopify's more than a million merchants coming next. That is four competing systems trying to solve the same problem, each backed by a company that wants its version to become the standard everyone else has to use.
In September 2026, Visa, Mastercard, and Ant International said they would work toward making their systems talk to each other. That sounds promising, but reporting on the announcement noted it arrived as a statement of intent, with no shared technical rules and no governing body yet. In plain terms, everyone agrees they should cooperate, but nobody has agreed on how.
There is a working template for this exact problem. Over the past few years, the tech industry replaced passwords with passkeys, a way to prove who you are using your phone instead of typing a password. More than one billion people have turned on a passkey, spread across over 13 billion online accounts. That only worked because companies agreed on one shared standard instead of each building their own.
The other lesson comes from the credit card chip. Card companies knew for years that the magnetic stripe was unsafe, but only fixed it after US fraud losses passed 16 billion dollars in a single year. Once chip cards became standard, in-person fraud dropped by 70 percent or more in country after country. The fix worked. It just took a decade of losses to force it.
If you sell anything online, AI shopping agents are already showing up in your traffic whether you built for them or not. The smart move is not to wait for one perfect standard to win. Start testing agent checkout on platforms that already carry real volume, such as Shopify or Etsy through ChatGPT, while keeping firm spending limits and manual review above whatever dollar amount matters to your business. The standard will eventually consolidate. Businesses that show up early with sensible guardrails will have a head start once it does.