The payment market for artificial intelligence is rapidly segmenting, and in my firm belief, traditional card systems are not a one-size-fits-all solution here. A joint analysis by Visa specialists and the blockchain analytics company Artemis clearly distinguishes between two fundamentally different types of AI agent transactions.

Two worlds of payments: purchases for humans and settlements between machines

The first scenario involves operations we are familiar with, carried out on behalf of a person: booking tickets, managing subscriptions, ordering goods. Here, as rightly noted, cards remain a convenient and competitive tool, as the amounts are comparable to regular retail trade.

However, the second scenario is the "circulatory system" of the future agent economy: frequent microtransactions for amounts under $1. This involves paying for a single API request, purchasing computing power, or accessing a software tool. It is precisely here, and I fully agree with this, that the fixed fees of card networks make them economically unviable.

As an alternative, experts point to blockchains with low fees, such as Solana and Base, where the transaction cost is fractions of a cent. The report examines in detail two open protocols: x402 (active on Base, Solana, and Polygon) and the Machine Payment Protocol (MPP), which supports both blockchain settlements and traditional fiat channels.

Real numbers and a hybrid future

According to Artemis data as of April 21, after rigorous filtering of test and dummy transactions, approximately $15 million (109.6 million transactions) passed through the x402 protocol, while MPP handled only $25,000 (115,000 transfers). It is important to emphasize: the raw data looked much more impressive, but after cleaning out the "junk," the picture became objective. This further proves that the market is still in its infancy.

The authors predict a hybrid model: cards will be used for authorization and large purchases on human command, while stablecoins will become the primary means for "seamless" settlements between programs. Both methods could even be applied within a single task at different stages.

The main stumbling block — trust and responsibility

The key unresolved issue, according to the analysts, remains the question of responsibility. Who will pay 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 were designed for humans and are not equipped to handle thousands of automatic transactions per hour.

An additional risk factor is malicious instructions that could redirect a payment or initiate an unplanned purchase. Until this legal and technical "gray area" is resolved, mass adoption will remain in question.

Analyst's opinion from Cryptalist: The payment market for AI agents is not just an evolution, but a tectonic shift. Stablecoins indeed appear to be a natural solution for micropayments here, but without solving the trust problem and establishing clear rules of the game for all participants (from the developer to the card issuer), this potential will remain unrealized. The current figure of $15 million is just the tip of the iceberg, but turning it into a full-fledged industry will require years and the joint efforts of regulators and the crypto industry.