๐ค Your AI assistant helps you buy thingsโsomeone finally handles the payment step
Last week in Hangzhou, something happened that didnโt make the hot news at the time, but may end up in business school textbooks:
A small-to-mid-sized export trading factory used an AI agent to complete the entire workflow of requesting quotes, placing orders, and making payments to overseas suppliers automatically.
No human needs to click a confirmation buttonโno copy-paste of card numbersโno waiting for finance approvals.
This became the Greater China regionโs first real-world B2B AI agent transactionโcompleted jointly by Visa ร LianLian International ร LoopXPay, launched in Hangzhou on July 24.
---
๐ณ Where did AI shopping usually get stuck at the end?
You may have heard that AI can help you book flights, write reports, or even make investment decisions. But thereโs one step thatโs been hard to break through: payment authorization.
The AI picks the items, then you hit payโan input screen for a bank card pops up. The AI canโt read it, and the user has to do it themselves. This step interrupts the entire automation chain.
The reason is simple: the machine has no proof that itโs legally allowed to spend money.
---
๐ How did Visa solve this problem?
Visa introduced the โAgentic Directoryโ (a Trusted Agent Directory), essentially giving AI a work badge.
In this directory, AI agents registered there have a clear identity and defined authorization scopeโwhat company they represent, how much they can spend, and what types of goods they can buy.
Merchants can also see which AI is sent and decide whether to accept the order.
Itโs like real-world corporate procurement authorization: salespeople have limits on how much they can sign for, and finance can understand it at a glance.
---
๐ Why choose Hangzhou this time?
China is the worldโs most densely concentrated manufacturing supply-chain region. Countless small factories process cross-border procurement every dayโraw materials, components, and packagingโwith suppliers across Southeast Asia, Japan and South Korea, and Europe.
Traditional process: procurement โ price comparison โ placing orders โ finance making the transferโeverything depends on people running it. A single payment process taking 3 days is the norm.
After AI agents get involved: from quote request to payment, in theory it can be compressed to minutes.
---
๐ก Behind this trend is a battle to standardize payment protocols
For AI agents to pay autonomously, they need a common โlanguage.โ Currently, two main approaches are racing:
โ x402 protocolโdominated by the Ethereum ecosystem, with the highest transaction volume. But Visa and Artemis reports point out that over 90% of x402 transactions are spam/volume tests, and the actual real market is about $15 million.
โก MPP Machine Payment Protocolโjointly introduced by Stripe and Tempo. It supports stablecoins + bank cards. In July, Visa and Mastercard joined the ecosystem, and the number of integrated services is growing rapidly.
Both approaches are competing: whoever captures the first wave of real transaction traffic first may become the standard for the next generation of payments.
---
๐ฌ A thought-provoking question:
If an AI agent can spend the companyโs money on its ownโhow do managers control it? How does finance review it?
If this isnโt resolved, AI payments can only stay in the experimental stage. Visaโs โTrusted Agent Directoryโ is just the first attempt, but the real compliance framework is still on the way.
How far do you think AI autonomous payments are from your company? Letโs discuss in the comments ๐
#AIๆฏไป #ๆบๅจๆฏไป #Tempo #blockchain