Visa report: AI agent payments have entered the practical stage, and stablecoins are more suitable for high-frequency small payments

Visa and Artemis jointly released a research report, using on-chain real-world testing data for the first time to prove that AI agent payments have moved beyond the “concept validation” stage and into true commercial scenarios where transactions can actually run. Visa and data analytics firm Artemis jointly released a report, using on-chain data from two major protocols, x402 and MPP, to verify that AI agent payments have moved from theory to practice, processing more than 100 million transactions within the year.

(Backgrounder: Cloudflare choosing Coinbase vs Stripe? That vote decides the AI agent payment standard) (Background supplement: Stripe launches AI Agent fully automated payment testing—supports USDC payments on Base via x402)

Table of contents

Toggle

  • AI agent payments: not just “replacing humans to swipe cards”
  • Why small payments need a new track
  • Credit cards and stablecoins aren’t “either/or”
  • Trust issues: who is responsible if an agent pays the wrong amount

Table of contents

Toggle

  • AI agent payments: not just “replacing humans to swipe cards”
  • Why small payments need a new track
  • Credit cards and stablecoins aren’t “either/or”
  • Trust issues: who is responsible if an agent pays the wrong amount

AI agent payments: not just “replacing humans to swipe cards”

The report categorizes AI agent commerce into two major types: macro commerce and micro commerce.

In macro commerce, AI agents represent humans to carry out consumption behaviors such as booking flights and managing subscriptions. The payment amounts fall within the range familiar to humans, and existing credit card rails can be used. Micro commerce is a different story—high-frequency, low-value payments between software systems, typically under $1, and sometimes only a few cents.

The report notes that MPP (built by Stripe and Tempo, with Visa also contributing), which went live in mid-March 2026, had already processed about 115k transactions in the first few weeks before it started running, with cumulative settlement of about $25k. The x402 protocol, launched earlier in May 2025, performed even more remarkably: handling about 109.6 million transactions, with an adjusted transaction volume of about $15 million, mainly concentrated on Base, Solana, and Polygon[1].

Why small payments need a new track

The report explains the problem with a simple comparison: the average payment per transaction under the two protocols is only a few cents. If you use traditional credit cards, the fixed per-transaction fees can be higher than the payment itself, so the economic model simply cannot work.

Visa’s data team says that what truly changes the landscape comes from two things happening at the same time: AI agents with budget-management capabilities create stable demand, and a new batch of blockchains pushes settlement costs down to the decimal level. With both combined, payments from $0.01 to $1 become feasible for the first time.

The report also mentions that the x402 Foundation, established just last July, already has 40 members, including organizations such as AWS, Google, Visa, Mastercard, Stripe, Coinbase, and Circle—indicating that the industry is accelerating toward unified standards[2].

Credit cards and stablecoins aren’t “either/or”

The report’s conclusion is clear: the future will not be a zero-sum game where “credit cards replace stablecoins” or “stablecoins replace credit cards.” Credit cards fit AI agent spending at the macro level, while stablecoins are better suited for machine-to-machine micro payments. In real scenarios, both may appear at different stages of the same task.

For the Taiwan market, the logic of AI agent payments applies similarly. Taiwan’s e-commerce transactions are dominated by mobile payments. If AI agents can autonomously handle small API calls and cloud computing billing, the stablecoin advantage of “a few cents per transaction” will be even more evident. Currently, Taiwan’s banking industry is actively testing stablecoin settlement. If, in the future, AI agents automatically pay for logistics, inventory management, or small cross-border purchases, a dual-rail model combining credit cards and stablecoins will likely be the most probable path to adoption.

Trust issues: who is responsible if an agent pays the wrong amount

The report also points out the biggest blind spot of AI agent payments: responsibility attribution. When an agent buys the wrong thing or has its budget redirected through malicious prompt instructions, who should take the blame—who is tasked with delivering, the platform that executes the agent, the company that trains the model, or the merchant that receives the funds?

Existing refund rules and dispute-resolution frameworks were designed for humans. When agents trade at a frequency of thousands of times per hour, and funds flow among multiple agents, the difficulty of tracking mistaken payments increases exponentially.

V-0.22%
NET-3.08%
COIN3.53%
SOL-1.49%
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned