The payment ecosystem for artificial intelligence is beginning to take clear shape. A joint analysis by Visa experts and the blockchain analytics company Artemis shows that traditional cards and stablecoins will occupy fundamentally different niches in the sphere of transactions initiated by AI agents.

Two Worlds — Two Payment Paradigms

The researchers divided all AI agent operations into two distinct types. The first is purchases made on direct human instruction: booking tickets, managing subscriptions, ordering goods. Here, transactions are essentially no different from regular online commerce, and Visa with cards remain the dominant tool.

The second type is micropayments between programs: frequent transfers of amounts under $1 for access to data, computing power, or individual API requests. It is here, in my observation, that the real breakthrough lies. The fixed fees of card infrastructure make it economically unviable for such operations. At the same time, in networks like Base and Solana, the transaction cost is fractions of a cent, which is ideal for high-frequency micro-settlements.

Protocols and Real Numbers

The report examines in detail two open payment protocols: x402 (active on Base, Solana, and Polygon) and the Machine Payment Protocol (MPP), which supports settlements both through blockchains and traditional systems. Artemis data as of April 21 shows that after filtering out fake and test operations, approximately $15 million (109.6 million transactions) passed through x402, and about $25,000 (115,000 transfers) through MPP.

It is important to note that the raw data was significantly higher: $135.7 million and 178.3 million operations for x402. The difference is colossal — 9 times. This once again confirms how critical filtering out "junk" transactions is for an objective assessment of the AI payments market.

The Main Stumbling Block — Trust

The key unresolved issue highlighted by the authors is the question of responsibility. Who should be held accountable for an erroneous or unauthorized purchase made by an AI agent? The owner, the platform operator, the model developer, or the seller? Existing chargeback mechanisms are designed for human actions and are completely unsuited for disputing thousands of automatic transactions per hour.

An additional risk is malicious instructions capable of redirecting a payment or forcing the agent to make an unplanned purchase. Without resolving these issues, the mass adoption of AI payments will remain in question.

Expert Opinion

A hybrid model, where cards are used for authorization and large purchases, and stablecoins for micropayments between programs, appears to be the most rational scenario. However, it is precisely the issue of trust and security that will become the bottleneck determining when this model moves beyond experiments. The market is clearly moving toward a division of labor, and stablecoins are gaining a clearly defined niche here.