Cards and stablecoins will occupy different niches in AI agent payments, according to a report by Visa and blockchain analytics firm Artemis.
The authors identified two categories of such transactions:
- human-directed purchases: booking travel, managing subscriptions, and ordering goods. These transactions are similar to regular online commerce and can be processed through payment cards;
- frequent transfers of less than $1 between programs. For example, an AI agent might pay for individual requests to online services, data retrieval, computing power, or software tools.
Visa and Artemis estimate that fixed fees make card infrastructure unsuitable for such micropayments. In Base and Solana, transaction fees are fractions of a cent.
The report examines two open payment protocols: x402 (primarily active on Base, Solana, and Polygon) and the Machine Payment Protocol (MPP). The latter supports settlements via blockchains and traditional payment systems.
The authors suggest a hybrid model where cards are used for authorization and human-directed purchases, while stablecoins are used for program-to-program settlements. Both payment methods can be applied at different stages of a single task.
According to Artemis data as of April 21, excluding identified fake and test operations, approximately $15 million was processed through x402 across 109.6 million transactions. For MPP, the figures were about $25,000 and 115,000 transfers, respectively.
The initial figures were significantly higher: $135.7 million and 178.3 million operations via x402, and over $38,000 and about 184,600 transactions via MPP. The difference is due to the exclusion of operations Artemis classified as fake activity or testing. The figures were calculated by the company, and no independent assessment is provided in the report.
The authors identified trust as the main unresolved issue. It remains unclear who should be responsible for erroneous or unauthorized purchases: the AI agent owner, the platform operator, the model developer, or the vendor.
Existing chargeback rules are designed for human transactions. There is no established mechanism for disputing thousands of automated transactions per hour. Malicious instructions that can redirect payments or cause an AI agent to make an unplanned purchase add additional risk.
In March, Artemis analyst excluded fake operations from x402 statistics. After filtering, the transaction volume of AI agents was 15 times lower than Bloomberg’s initial estimate.
In June, Alchemy and Visa launched the AgentCard service for purchases through AI agents within user-set limits.
