The payment industry is on the verge of a fundamental split: traditional cards and stablecoins will serve fundamentally different segments of automated settlements. Experts at Visa, together with blockchain analytics company Artemis, conducted an in-depth study of AI agent payment flows and reached important conclusions.

Analysts identified two key categories of transactions performed by artificial intelligence. The first is purchases made on direct human instruction: booking tickets, managing subscriptions, ordering goods. In essence, these operations are identical to regular e-commerce and can be efficiently processed through traditional payment cards.

The second category is micropayments between programs, where amounts regularly drop below $1. This includes paying for individual API requests, access to computing power, purchasing data, or using software tools. It is here, in my estimation, that the main growth potential for cryptocurrency infrastructure lies.

Why stablecoins beat cards in micropayments

Fixed fees charged by card networks make traditional infrastructure economically unfeasible for transactions worth a few cents. At the same time, blockchains like Base and Solana demonstrate fees at the level of fractions of a cent per transaction. The report examines two open protocols: x402, whose activity is concentrated on Base, Solana, and Polygon, as well as the Machine Payment Protocol (MPP), which supports settlements both through blockchains and traditional systems.

In my view, the most interesting is the hybrid model proposed by the authors: cards remain for authorization and large purchases on behalf of a human, while stablecoins handle settlements between programs. Moreover, both methods can be applied at different stages of executing a single task, paving the way for creating fully autonomous financial agents.

Real numbers and the trust problem

According to Artemis data as of April 21, after filtering out dummy and test operations, approximately $15 million passed through the x402 protocol with 109.6 million transactions. For MPP, the figures were roughly $25,000 and 115,000 transfers. The raw numbers were significantly higher: $135.7 million and 178.3 million operations via x402, as well as over $38,000 and 184,600 transactions via MPP. The difference clearly demonstrates how important it is to critically evaluate raw data in this area.

The main unresolved issue remains trust. Who bears responsibility for an erroneous or unauthorized purchase made by an AI agent? The agent's owner, the platform operator, the model developer, or the seller? Existing chargeback mechanisms are designed for human operations and are completely unsuited for disputing thousands of automatic transactions per hour. Additional risk comes from malicious instructions capable of redirecting payments or initiating unplanned purchases.

Let me remind you that in March, an Artemis analyst had already excluded dummy operations from x402 statistics, after which the real volume of AI agent transactions turned out to be 15 times lower than Bloomberg's initial estimates. And in June, Alchemy and Visa launched the AgentCard service for controlled purchases via AI.

Expert commentary: The AI agent micropayment market is not just a niche story but a potential driver for mass stablecoin adoption. However, without solving the issue of legal liability and creating transparent mechanisms for disputing transactions, this segment risks remaining an experimental playground. Investors should closely monitor the development of the x402 and MPP protocols — they could become the de facto standard for automatic settlements.