The world of payments for AI agents is gradually taking on a clear structure. As recent research shows, traditional cards and stablecoins will serve fundamentally different segments of this ecosystem. Experts from Visa and the analytical company Artemis conducted an in-depth analysis and identified two key areas of automated transactions.
The first category is purchases made on direct human instruction. This includes booking tickets, managing subscriptions, or ordering goods. Such operations are essentially identical to regular online commerce, and classic card infrastructure is quite suitable for them. The second category is micropayments between programs themselves. This refers to frequent transfers of amounts less than one dollar. AI agents may pay for individual requests to cloud services, for data access, for computing power, or for using software tools.
It is here that traditional cards face a serious barrier. Fixed fees make their use for such small transactions completely unprofitable. At the same time, blockchains like Solana and Base offer fees amounting to fractions of a cent. This makes them an ideal environment for the microtransaction economy.
Protocols and Hybrid Models
The report examines two open payment protocols in detail. The first is x402, whose main activity is concentrated on the Base, Solana, and Polygon networks. The second is the Machine Payment Protocol (MPP), which supports settlements both through blockchains and through traditional payment systems. The study's authors suggest that in the future we may see a hybrid model: cards will be used for authorization and large purchases on human instruction, while stablecoins will be used for instant settlements between AI agents. Moreover, both payment methods could be applied at different stages of executing a single task.
The statistics presented by Artemis as of April 21 are quite telling. After excluding identified fake and test transactions, approximately 109.6 million transactions worth about $15 million were processed through the x402 protocol. For MPP, the figures are more modest: about $25,000 and 115,000 transfers. Interestingly, the initial "raw" data was significantly higher: $135.7 million and 178.3 million operations via x402. The difference demonstrates how important it is to carefully filter data to get a real picture of activity.
The Main Problem is Trust
Despite technological readiness, trust remains a key unresolved issue. It is still unclear who should bear responsibility for an erroneous or unauthorized transaction: the owner of the AI agent, the platform operator, the model developer, or the seller. Existing chargeback mechanisms are designed for human actions. There is simply no established process for disputing thousands of automatic micropayments per hour. An additional risk is posed by malicious instructions capable of redirecting a payment or forcing the agent to make an unplanned purchase.
Recall that in March, Artemis analysts already drew attention to inflated estimates by Bloomberg, excluding fake transactions from x402 statistics. After filtering, the real volume of AI agent transactions turned out to be 15 times lower. And in June, Alchemy and Visa launched the AgentCard service, allowing purchases through AI agents within user-set limits.
Expert Opinion: The market for AI agent payments is at the stage of standard formation, and stablecoins here are not just an alternative, but the technologically only correct solution for microtransactions. However, until the issue of legal liability is resolved and clear mechanisms for disputing operations emerge, mass adoption will remain in question. It is the legal framework, not transaction speed, that will become the main driver or brake for this industry in 2024-2025.