The world of payments for artificial intelligence is splitting into two distinct camps, with stablecoins occupying a strategically important but narrow niche. In a deep analysis conducted by my team together with the blockchain analytics platform Artemis, we concluded that traditional cards and digital currencies will serve fundamentally different types of AI agent transactions.

We identified two key categories of operations. The first is purchases made on direct human instruction: booking tickets, managing subscriptions, ordering goods. These transactions are essentially identical to regular online commerce and can easily be processed through traditional payment cards. The second category is frequent, almost microscopic transfers between programs, with amounts rarely exceeding $1. This includes payments for individual requests to online services, data access, computing power rental, or the use of software tools.

This is where card infrastructure fails. Fixed payment processing fees make micropayments economically unviable. At the same time, blockchains such as Base and Solana offer fees measured in fractions of a cent. This makes stablecoins an ideal tool for machine-to-machine settlements.

Hybrid Model of the Future

During the research, we examined in detail two open payment protocols: x402, whose activity is concentrated on the Base, Solana, and Polygon networks, and the Machine Payment Protocol (MPP), which supports settlements both through blockchains and traditional systems. Our analysis shows that a hybrid model is most likely: cards will be used for authorization and large purchases on behalf of humans, while stablecoins will take on the role of the "circulatory system" for settlements between programs. Both methods can be applied at different stages of executing a single task.

According to our data as of April 21, after thorough filtering and excluding fake and test operations, approximately $15 million was processed through the x402 protocol across 109.6 million transactions. The MPP figures are more modest: roughly $25,000 and 115,000 transfers. The raw numbers were significantly higher ($135.7 million and 178.3 million operations for x402), but we deemed it necessary to clean the statistics of anomalies.

The Key Problem — Trust

The main unresolved issue remains trust and accountability. Who should be responsible for an erroneous or unauthorized purchase made by an AI agent: the owner, the platform operator, the model developer, or the seller? Existing chargeback rules are designed for human actions and are completely unsuitable for thousands of automatic transactions per hour. An additional risk is posed by malicious instructions capable of redirecting a payment or forcing the agent to make an unplanned purchase.

My expert opinion: Stablecoins find their "killer feature" not in replacing fiat money in retail, but in creating a new economic paradigm for machines. The problem of trust and legal liability is a stumbling block that will be resolved over years through smart contracts and insurance protocols. However, right now, when the transaction volumes of AI agents already number in the tens of millions of operations, we are witnessing the birth of the fundamental infrastructure for the web3 economy of the future.