Invest1 publisher3 min readPublished
TRM screens out $27 million of x402's $52.7 million before counting any agents
TRM Labs put 198.9 million x402 settlements through a filter for self-payments and thin sellers. $25.62 million survived. Agents got 0.6% to 7.5% of that, on a model TRM says may understate them.
The Investor · Invest desk
What happened
- TRM Labs published a report Wednesday examining $52.7 million across 198.9 million settlements handled by known x402 facilitators on Base, Solana and Polygon since May 2025.
- Screening out self-payments, bulk flows from one or two payers and sellers with fewer than 10 buyers left $25.62 million that TRM treated as likely commerce.
- Of that remainder, TRM estimated that between 0.6% and 7.5% of value appeared to come from AI agents.
- USDC settled 99.6% of the value over the whole period, or $52.47 million.
- Binance's Agent OS launch in August included an x402 payment layer, and Amazon announced AgentCore Payments with Coinbase and Stripe in May for agents paying with stablecoins.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint At the top of TRM's range, agent-driven value is 3.6% of everything x402 has settled, so a take-rate business on this rail has to be priced off that pool.
- cost Facilitators, not payers, carry the network fee on every one of the 198.9 million settlements they broadcast.
- exposure Because ownership declarations are voluntary and mostly unused, a counterparty to a disputed agent payment has no declared owner to chase.
- contradiction The estimate cannot settle the argument it will be used to settle: TRM says its models cannot conclusively separate agents from scripts, so sceptics and funders can both keep the position they came in with.
Average settled value per x402 payment is about 26.5 cents: $52.7 million spread over 198.9 million settlements [1][1]. TRM Labs built its agent filter for exactly that size, looking for facilitator-broadcast payments whose amounts varied and averaged under a dollar, with a stricter version requiring the pattern to hold across months plus either public agent registration or payments to multiple sellers [4].
The exclusions removed $27.08 million, or 51.4% of everything settled [2]. Coinbase launched x402 in 2025 to fold payment into the web request: the buyer gets a price, signs an authorisation, and a facilitator verifies it, submits the transaction and pays the network fee [7]. TRM's point is that a plain script can do all of that, and so can a scheduled job, a load test, or a payer paying itself [8].
Run TRM's percentages against what survived and agent-driven commerce since May 2025 comes to between $153,720 and $1.92 million [3]. Coinbase's Base has been recruiting agent and payments startups through a $1 million accelerator [11]. The accelerator is a little over half the top end of the measured market [4].
"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 said [5]. What the filter detects is shopping around [6]. It also said: "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" [6].
The composition moved over the period. Late in 2025 the flows included apparent meme-token minting and payments to one AI-analysis service; by early 2026 volume had concentrated in a single payment contract, and around midyear AI-service payments returned through an agent-payment router [9].
"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" [13]. The report's ask was specific: "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" [14].
I would underwrite the $25.62 million [2] and hold the agent share as unmeasured, since the test that yields 7.5% is a test for variety and production agents are narrow by design. The counter-case is TRM's own caveat read forward: if most working agents pay one service repeatedly, the real share sits above 7.5% and the capital going in is early [6]. A third possibility is that the number climbs with no change in behaviour at all, once enough addresses declare ownership in the registries TRM says most participants do not use [13].
What to watch
- Whether the agent-payment router that carried AI-service payments back onto x402 around midyear survives the concentration exclusions in TRM's next sample.
- Whether Amazon's AgentCore Payments produces multi-seller payment patterns that clear TRM's stricter agent test.
- Whether volume from Binance's Agent OS x402 layer appears in a later measurement, and under which facilitator.