Back to News
NewsStablecoin

Visa Assigns Stablecoins a Role in AI Agent Micropayments

Visa and Artemis report on the distinct roles of cards and stablecoins in AI agent payments, highlighting micropayments and trust issues.

Mercation Admin
July 16, 2026 2 min read
Visa Assigns Stablecoins a Role in AI Agent Micropayments

Visa, in collaboration with blockchain analytics firm Artemis, has concluded that cards and stablecoins will serve distinct roles in AI agent payment systems.

The study identified two main types of transactions involving AI agents:

  • Purchases made on behalf of individuals, such as booking travel, managing subscriptions, and ordering products, which resemble traditional online shopping and can utilize payment cards;
  • Frequent, low-value transfers below $1 between applications, where AI agents might pay for specific requests to online services, data access, computing resources, or software tools.

Visa and Artemis noted that fixed fees associated with card transactions render them unviable for micropayments, while platforms like Base and Solana offer transaction fees that are only fractions of a cent.

The report evaluated two open payment protocols: x402, which operates mainly on Base, Solana, and Polygon, and the Machine Payment Protocol (MPP), which facilitates transactions across both blockchain and traditional payment networks.

The authors suggested a hybrid approach where cards are used for authorizing and making purchases for individuals, while stablecoins are utilized for inter-program transactions. This dual method can be applied at various stages of a single task.

According to Artemis data from April 21, after excluding fraudulent and test transactions, the x402 protocol processed around $15 million across 109.6 million transactions, while the MPP recorded about $25,000 in 115,000 transfers. Initial figures were significantly higher, with $135.7 million and 178.3 million operations for x402, and over $38,000 with around 184,600 transactions for MPP. The revisions were made after filtering out fraudulent activities.

One of the significant challenges highlighted by the authors is the issue of trust regarding erroneous or unauthorized transactions. It remains unclear who should be held accountable—the AI agent's owner, the platform operator, the developer, or the vendor.

Current refund policies cater to human transactions, and there is no established process for disputing numerous automated transactions occurring every hour. This situation raises additional concerns, particularly regarding malicious commands that could misdirect payments or prompt the AI agent to execute unintended purchases.

In March, an Artemis analyst noted that once fraudulent transactions were excluded, the transaction volume attributed to AI agents was found to be 15 times lower than initial estimates from Bloomberg. Additionally, in June, Alchemy and Visa introduced the AgentCard service, allowing purchases through AI agents within user-defined limits.