x402 agent payments are real, but most of the volume is not agents. On September 9, 2026, blockchain-analytics firm TRM Labs published an on-chain analysis of x402, the protocol that lets software pay for a web resource. Of $52.7 million in settlements, it estimates that between 0.6% and 7.5% of the screened commerce appears to be agentic (TRM Labs).

If you run a store, this is a useful reality check on headline numbers for “agent payments”.

What TRM measured

TRM counted settlements through known x402 facilitators on Base, Solana and Polygon since May 2025: roughly $52.7 million across 198.9 million transactions.

It then applied three screens to isolate likely commerce:

  • remove addresses paying themselves;
  • remove bulk flows from one or two payers;
  • remove sellers with fewer than ten distinct buyers.

TRM says about half of settled volume falls away after these screens, leaving $25.62 million of likely commerce.

How it tells an agent from a script

It does not claim to. TRM states that “a scheduled script and an agent produce the same record”, so it bounds the answer with two tests:

  • Permissive: facilitator-broadcast payments, sub-dollar on average, with varying amounts rather than a fixed price.
  • Strict: the same pattern sustained across months, plus either registration in a public agent registry (ERC-8004) or payments to more than one seller.

Result: 0.6% to 7.5% of the screened commerce looks agentic. TRM flags the weakness itself: treating a repeated price as a script “may understate the space”, because many agents today could be single-purpose.

What the money was actually buying

TRM’s category mix is not retail. It describes late-2025 volume as speculation, much of it consistent with meme-token minting, then a single dominant payment contract in the first half of 2026. AI and agent services returned in mid-2026 through an agent-payment router. It lists AI services, financial services, data and developer APIs and token mints as the merchant groups, and no consumer goods.

Nearly all of it settles in one asset: $52.47 million of $52.68 million (99.6%) in USDC. TRM puts the monthly volume moved by agents at roughly $5,000 to $11,000.

What this means for a merchant

  • Do not read protocol totals as shopper demand. TRM’s own advice is that the headline volume of an agent-payment channel is mostly not customers.
  • x402 is a rail for per-call machine services, not a checkout. Our x402 explainer covers where it sits next to ACP and UCP.
  • Attribution is the open problem. TRM notes that ownership claims in agent registries are voluntary and “currently not utilized by the majority of participants”, and that no human approves the payment.
  • What attracts agent buyers is design. TRM’s wording: it needs to be machine-readable and have per-call pricing on an endpoint a buyer can discover without a human ever seeing the page.

FAQ

Are AI agents actually paying with x402?

Some are, but they are a small share. TRM Labs estimates that 0.6% to 7.5% of $25.62 million of screened x402 commerce looks agent-originated, using a permissive and a strict test. TRM also says it cannot read agency directly from a transaction.

Can on-chain data tell an AI agent from a script?

Not reliably. TRM says scheduled jobs, load tests, self-dealing and ordinary automation leave the same on-chain record as an agent. Voluntary agent registries such as ERC-8004 help only for the minority of participants who use them.

Should my store accept x402 payments?

TRM’s data shows x402 volume is mostly AI services, data APIs, financial services and token mints, not retail goods. For a store selling physical products, the checkout path agents use is a different problem: product data, cart and a purchase step an agent can complete.

Sources

The rail can work while the store behind it is still unbuyable by an agent: that gap, not the payment protocol, is what AgentReady audits.