Most of the payment volume moving through Coinbase’s x402 protocol does not appear to come from AI agents, according to blockchain intelligence firm TRM Labs.
In a report published Wednesday, TRM examined $52.7 million across 198.9 million settlements processed by known x402 facilitators on Base, Solana, and Polygon since May 2025. After removing self-payments and other anomalous flows, the researchers estimated that only 0.6% to 7.5% of the remaining commerce by value appeared to come from AI agents.
TRM wrote, “The assumption is that a true agent explores across multiple services and products, while an address repeating the same price behaves more like a script hitting one service over and over. This is a deliberate modeling choice, and it may understate the space: many agents today could be single-purpose, paying one service repeatedly, which this test would read as a script.”
How x402 works
Coinbase launched x402 in 2025 as a way to integrate payments directly into web requests. A buyer receives a price quote and signs a payment authorization. A facilitator then verifies that authorization, submits the blockchain transaction, and pays the network fee.
According to TRM Labs, an ordinary script can complete that same sequence without any AI agent involved. Scheduled jobs, load tests, and self-dealing can also leave behind the same kind of on-chain footprint, which means aggregate protocol volume is not enough to measure agent commerce.
TRM’s filtering process
TRM removed self-payments, bulk flows from one or two payers, and sellers with fewer than 10 buyers. That left $25.62 million in activity the firm classified as likely commerce.
From there, the researchers screened for facilitator-broadcast payments with varying amounts averaging below $1. A stricter version of the test required that pattern to appear across multiple months, along with either public agent registration or payments to multiple sellers. TRM said those criteria could still miss genuine agents that repeatedly purchase the same service.
Activity patterns and token usage
Late-2025 activity included apparent meme-token minting and payments to one AI-analysis service, the report said. In early 2026, volume became concentrated in a single payment contract. Around midyear, payments for AI services returned through an agent-payment router.
Across the full period, USDC represented 99.6% of settled value, equal to $52.47 million.
Companies are still building around agent payments
Even as TRM questions how much current on-chain volume reflects real autonomous agent spending, companies have continued to invest in the category.
Binance’s Agent OS launch in August included an x402 payment layer. Coinbase used its $1 million accelerator to target agent and payments startups. In May, Amazon announced AgentCore Payments with Coinbase and Stripe, allowing agents to pay for online services with stablecoins.
Responsibility and compliance remain unresolved
Beyond usage measurement, TRM said it found gaps in assigning responsibility for agent payments.
“On-chain agent registries let individuals declare ownership of an agent address,” TRM Labs wrote. “However, this declaration is voluntary and currently not utilized by the majority of participants.”
TRM called for more accurate registration, counterparty reputation information that agents can check, and monitoring systems designed for high volumes of small payments.
“The rail already works. What is needed is accurate registration, counterparty reputation an agent can check on its own, and monitoring built for volume rather than value,” the report said. “Agentic commerce will need agentic compliance.”