198.9 million x402 payments look like an agent economy. TRM Labs says most of them aren’t.

TRM Labs analyzed roughly $52.7 million across 198.9 million x402 settlement transactions and found that headline protocol activity substantially overstates identifiable AI-agent commerce. The measurement problem is becoming attribution, not transaction count.

The signal

TRM Labs analyzed approximately $52.7 million across 198.9 million x402 settlement transactions on Base, Solana and Polygon. After removing self-payment and anomalous patterns, TRM estimated roughly $25.62 million as likely commerce. Its attribution methods then placed plausible AI-agent commerce at only a small fraction of that value.

Why this matters more than another x402 milestone

Our earlier x402 article explicitly refused to equate protocol activity with autonomous-agent commerce. TRM attempts to quantify that distinction and finds that most visible x402 settlement activity cannot confidently be described as agentic.

What TRM can—and cannot—see

On-chain payments do not carry a reliable label saying whether the payer was an LLM-driven agent, a scheduled script, ordinary automation or a human-controlled process. TRM therefore relies on screens and attribution heuristics. Its 0.6%–7.5% estimate should be treated as a range, not ground truth.

The stablecoin signal

TRM reports that 99.6% of x402 value in its dataset settled in USDC. Even if identifiable agent value remains small, the payment pattern is already distinct: high transaction counts, extremely low individual values and no human approval between machine decision and settlement.

What we watch next

Verified agent identity, unique agents, recurring machine buyers, seller concentration, paid-service categories, monthly attributable agent volume and whether new identity standards make autonomous-payment measurement materially cleaner.