TRM Labs Data Check: AI Agents Behind Under 8 Percent of x402 Payment Activity

TRM Labs has delivered the most sobering dataset yet on the agentic commerce narrative, estimating that AI agents account for somewhere between 0.6 percent and 7.5 percent of the value flowing through Coinbase’s x402 payment protocol after anomalous flows are stripped out.

The blockchain intelligence firm examined roughly 52.7 million USD across 198.9 million x402 settlements processed by known facilitators on Base, Solana, and Polygon since May 2025. After screening out self-payments, bulk flows from one or two payers, and sellers with fewer than ten buyers, researchers were left with 25.62 million USD in what they classified as likely commerce. Within that filtered pool, the share attributable to AI agents remained remarkably small under every model TRM tested.

Why measuring agent payments is genuinely hard

The core problem, according to the report, is that x402’s payment flow does not require intelligence to execute. Launched by Coinbase in 2025, the protocol integrates payments directly into web requests: a buyer receives a price quote, signs a payment authorization, and a facilitator verifies it, submits the blockchain transaction, and covers the network fee. An ordinary script can follow that exact sequence without any AI involved. Scheduled jobs, load tests, and self-dealing wallets can all produce blockchain records indistinguishable from genuine machine commerce at the transaction level.

TRM’s approach was behavioral rather than technical. “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,” the report explained. Researchers screened for facilitator-broadcast payments with varying amounts averaging under 1 USD, then applied a stricter test requiring that pattern to persist across months, combined with public agent registration or payments to multiple distinct sellers.

The firm was candid about the limits of its own methodology. Many deployed agents today are single-purpose tools that buy the same service repeatedly, which the exploration test would misclassify as a script. The modeling choice, TRM noted, “may understate the space.”

What the activity actually looked like

The on-chain history behind the numbers tells a story of experimentation rather than an autonomous economy taking root. In late 2025, activity included apparent meme-token minting and recurring payments to a single AI-analysis service. Volume then concentrated in one payment contract in early 2026, before AI-service payments returned through an agent-payment router around the middle of the year.

Across the entire period, USDC dominated settlement almost completely, accounting for 99.6 percent of settled value, or 52.47 million USD. The preference reflects the stablecoin’s role as the default denomination for machine-to-machine micropayments, where price stability matters more than speculative upside.

Industry keeps building ahead of demand

The measured reality has not slowed investment in agent payment infrastructure. Binance’s August Agent OS launch included an x402 payment layer, while Coinbase’s Base has targeted agent and payments startups through a 1 million USD accelerator program. In May, Amazon announced AgentCore Payments with Coinbase and Stripe, enabling AI agents to pay for online services using stablecoins.

That gap between infrastructure investment and observed usage is precisely why TRM framed its research as a calibration exercise. If agents are to transact at scale, the industry needs baselines that separate genuine autonomous commerce from scripted noise, otherwise adoption metrics become self-congratulatory fiction.

Compliance is the missing rail

Beyond measurement, TRM identified a more structural gap: assigning responsibility for what agents do. On-chain agent registries exist and let individuals declare ownership of an agent address, but the declaration is voluntary “and currently not utilized by the majority of participants.”

The firm called for three building blocks: accurate registration of agent ownership, counterparty reputation information that agents can check autonomously before transacting, and monitoring systems designed for high volumes of tiny payments rather than low volumes of large ones.

“The rail already works,” TRM concluded. “What is needed is accurate registration, counterparty reputation an agent can check on its own, and monitoring built for volume rather than value. Agentic commerce will need agentic compliance.”

A reality check, not a eulogy

The report lands at a moment when agentic commerce has become one of the most heavily marketed narratives in crypto, with x402 positioned as the native payment layer for an anticipated explosion of machine-to-machine transactions. TRM’s data suggests the explosion, if it comes, has not started yet.

For developers and investors, the takeaway is twofold. First, protocol transaction counts are a poor proxy for agent adoption, since scripts and self-dealing can inflate totals dramatically. Second, the compliance layer may prove as decisive as the payment layer, because enterprises will not delegate spending authority to autonomous wallets that cannot be attributed, audited, or held accountable.

The 0.6 to 7.5 percent range is wide, and TRM admits even the upper bound could be conservative. But until registration becomes standard and behavioral analytics mature, the honest answer to how much AI agents are actually spending online remains: far less than the hype suggests, and nobody can yet say precisely how much.