AI agent transactions much lower than reported after data cleanup

AI agent payments totaled about $1.6 million over the past 30 days after suspected wash trades were filtered out, according to Andreessen Horowitz partner Noah Levine. The figure is far below a widely referenced $24 million tally for a similar period.

Levine, in an X post on March 11, cited Allium Labs data indicating roughly $3 million in AI agent transactions over the most recent 30-day period, which falls to about $1.6 million after removing suspected wash trading. He contrasted that with a $24 million estimate based on data from x402.org. “The gap tells you how early-stage even the measurement infrastructure is,” Levine wrote.

Levine noted that most current activity centers on developer-focused services. Examples include Firecrawl, which sells web scraping for about one cent per query, Browserbase, which offers browser sessions built for AI, and Freepik, which provides AI image generation. While these services accept traditional cards, x402 allows one-off, pay-per-use access without subscriptions.

Levine pointed to integrations of x402 by Stripe, Cloudflare and Vercel, and indicated that Google has embedded the system into its agent payments protocol. He described most transactions as semi-autonomous, with humans still in the loop through platforms like Claude Code and the OpenClaw personal AI assistant.

On March 11, Coinbase announced that the x402 Facilitator now supports the Ethereum layer-2 network Polygon, enabling developers to accept USDC via x402 on Polygon in addition to Base and Solana. The company framed the update as a way to improve speed and lower fees for automated, machine-to-machine payments.

x402 is a payment standard developed by Coinbase that allows software agents to initiate and complete online payments programmatically. The standard removes manual checkout steps and supports pay-per-use access to APIs and web tools. The x402 Facilitator is a service layer that routes and settles these payments across supported networks and assets, with USDC used for stable pricing.

According to a study by the Bitcoin Policy Institute, AI models tend to favor Bitcoin and stablecoins over fiat currencies. Researchers tested 36 frontier AI models across 9,072 monetary scenarios. Overall, the models chose Bitcoin 48.3% of the time and selected it in 79.1% of multi-year store-of-value questions. Stablecoins dominated payment-related tasks, leading with 53.2% compared with Bitcoin’s 36%. Across substantive responses, 90.8% favored digitally native assets over fiat. The report also cited a global survey showing that about 35% of freelancers’ and online sellers’ annual earnings are paid in stablecoins, with nearly three-quarters saying stablecoin payments improved their ability to work internationally.