The precise volume of AI agent transactions on blockchain protocols remains difficult to ascertain, with current analysis methods potentially understating their activity due to challenges in distinguishing automated scripts from genuine AI agents. Research by TRM Labs examining payment volume on Coinbase’s x402 protocol suggests that AI agents account for a small percentage of transactions, but the methodology itself highlights significant limitations in accurately identifying and attributing such activity.
TRM Labs analyzed $52.7 million in transactions across 198.9 million settlements processed by known x402 facilitators on the Base, Solana, and Polygon blockchains since May 2025. After excluding self-payments, bulk flows from a limited number of payers, and sellers with fewer than 10 buyers, the firm was left with $25.62 million in likely commerce. Within this refined dataset, TRM Labs estimated that only 0.6% to 7.5% of the commerce value originated from AI agents. USDC accounted for 99.6% of the settled value, totaling $52.47 million.
Technical Hurdles in Distinguishing Agents
The report, published Wednesday, detailed the technical hurdles in differentiating AI agents from other automated processes. According to TRM Labs, a true AI agent is expected to explore multiple services and products, whereas an address repeatedly interacting with the same price point could behave more like a script. This distinction is a deliberate modeling choice, but TRM cautioned that it may understate the space, as many agents today could be single-purpose and repeatedly pay for the same service, which this test would interpret as a script.
Furthermore, the analysis noted that scheduled jobs, load tests, and self-dealing can produce similar blockchain records, making protocol totals insufficient for accurately measuring agent commerce. The report also pointed out that on-chain agent registries, where individuals can declare ownership of an agent address, are voluntary and currently not utilized by the majority of participants, further complicating attribution.
The x402 Protocol and Evolving Infrastructure
The x402 protocol, launched by Coinbase in 2025, integrates payments into web requests, where a buyer receives a price, authorizes payment, and a facilitator verifies and submits the transaction, paying the network fee. While the underlying infrastructure for agent payments is functional, TRM Labs emphasized the need for improvements. The report called for more accurate registration, counterparty reputation information that agents can check independently, and monitoring systems built for high volumes of small payments rather than just high-value transactions.
Despite these analytical challenges, the development of agent payments continues. Binance launched Agent OS with an x402 payment layer in August, and Coinbase’s Base has targeted agent and payments startups through its accelerator program. In May, Amazon announced AgentCore Payments in collaboration with Coinbase and Stripe.
The TRM Labs analysis underscores that while agent-based commerce is developing, the current methods for tracking and verifying AI agent activity are nascent and require significant enhancement. The report implies that accurate registration, robust reputation systems, and tailored monitoring are essential for the future of agentic commerce and its compliance.
Why This Matters
The materials describe a narrow update: TRM Labs analyzed $52. The exact percentage of payment volume attributable to AI agents.
Broader Context
Source materials place the factual news in this context: Most payment volume on Coinbase’s x402 protocol isn’t coming from AI agents, according to blockchain intelligence company TRM Labs.
Leave a Reply