TRM: AI agents account for just 0.6%–7.5% of x402 “likely commerce”
The study reviewed 198.9 million settlements since May 2025 and found USDC made up 99.6% of settled value.
TRM Labs put a number on the “agentic commerce” narrative around Coinbase’s x402 payments rail and found it remains a small slice of measurable activity. After filtering out self-payments and other anomalous flows, TRM estimated only 0.6%–7.5% of remaining commerce by value appeared to come from AI agents under its models.
Key Takeaways
- Known x402 facilitators processed about 198.9 million settlements totaling roughly $52.7 million on Base, Solana, and Polygon since May 2025 in TRM’s examination window.
- Filtering out self-payments and other anomalous flows left $25.62 million in what TRM labeled “likely commerce.”
- Under TRM’s models, only 0.6%–7.5% of that remaining commerce by value appeared attributable to AI agents.
- USDC represented 99.6% of x402 settled value over the period, or about $52.47 million.
TRM Puts Numbers on x402’s “Agentic Commerce” — and It’s Small
TRM Labs’ latest look at Coinbase’s x402 payment protocol lands on an awkward conclusion for the AI-agent trade: there is real settlement activity, but very little of it can be cleanly labeled “agents spending money” once obvious noise is stripped out.
Across x402 settlements processed by known facilitators on Base, Solana, and Polygon since May 2025, TRM examined roughly $52.7 million in value across 198.9 million settlements. After excluding self-payments and other anomalous flows, TRM estimated $25.62 million in “likely commerce,” then modeled how much of that commerce looked agent-driven. The range it published was 0.6%–7.5% of the remaining commerce by value.
For traders, the point is not that x402 is “dead.” The point is that the popular proxy claim, that AI agents are already driving meaningful on-chain payments, is not supported by what can be inferred from settlement traces alone, even on a rail explicitly pitched for web-native payments.
The Dataset: 198.9M Settlements, $52.7M Value, and USDC’s 99.6% Dominance
x402 is a Coinbase-launched (2025) protocol that embeds payments into standard web requests. The buyer receives a price and signs a payment authorization. A facilitator, meaning a service that verifies the authorization, submits the on-chain transaction, and pays the network fee, handles the blockchain leg.
TRM’s dataset covered settlements processed by known facilitators on three chains: Base, Solana, and Polygon. The headline totals were large in count and modest in dollars: 198.9 million settlements and about $52.7 million in settled value since May 2025.
The composition matters as much as the totals. USDC accounted for 99.6% of settled value across the full period, or $52.47 million. In practice, that makes the current x402 footprint a USDC payments story first, and an “agent payments” story only if attribution improves.
TRM also described how activity shifted over time inside that window. Late 2025 included apparent meme-token minting and payments to one AI-analysis service. Early 2026 saw volume concentrate in a single payment contract. Around mid-2026, TRM said AI-service payments returned through an agent-payment router.
Why On-Chain Patterns Can’t Prove an Address Is an AI Agent
TRM’s core warning is methodological: the chain does not tell you whether a payment was initiated by an AI agent, a simple script, or operational plumbing. The x402 flow is deterministic enough that “agent-like” behavior can be mimicked by ordinary scripts, scheduled jobs, load tests, and self-dealing, producing similar on-chain records.
That ambiguity is why TRM treated “agent payment volume” as a modeled estimate rather than a directly observable metric. Its report framed a key behavioral heuristic this way: “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.”
TRM also made explicit that this choice can cut the other way. “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.”
The filtering pipeline matters because it is doing most of the work. TRM removed self-payments, bulk flows from one or two payers, and sellers with fewer than 10 buyers to arrive at $25.62 million in “likely commerce.” It then screened for facilitator-broadcast payments with varying amounts averaging under $1. A stricter test required that pattern across months, plus public agent registration or payments to multiple sellers. TRM cautioned that these criteria could miss genuine agents that repeatedly buy the same service.
The practical takeaway is that the 0.6%–7.5% range is not a “true” number waiting to be discovered onchain. It is the output of assumptions about what agents should look like when they pay, and those assumptions are still contested.
Identity, Reputation, and “Agentic Compliance”: The Missing Layer for Machine Payments
TRM’s forward-looking argument is that the rail is ahead of the accountability layer. x402 can settle high-frequency payments. The missing piece is assigning responsibility and risk when the payer is software.
TRM pointed to weak adoption of voluntary identity tooling. “On-chain agent registries let individuals declare ownership of an agent address,” it wrote. “However, this declaration is voluntary and currently not utilized by the majority of participants.”
The fixes TRM called for are straightforward but non-trivial to operationalize at scale: accurate registration, counterparty reputation information that agents can check on their own, and monitoring designed for volume rather than value. “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,” TRM wrote, adding: “Agentic commerce will need agentic compliance.”
What moves this story from narrative to mechanism is whether major facilitators and Coinbase-linked teams start standardizing identity and attestation, rather than leaving registries voluntary. The other hard signal is growth quality: whether x402 settlement value and TRM’s “likely commerce” subset expand materially beyond the $52.7 million total and $25.62 million filtered slice, and whether that growth is distributed across multiple sellers instead of concentrated flows. A third tell is payment-asset diversity. USDC at 99.6% is efficient, but a shift away from that share would indicate experimentation in how x402 is used, not just that it exists.
My Read: The Trade Isn’t ‘Agents Are Here’ — It’s ‘Payments Rails Are Here, Attribution Isn’t’
The threshold that matters here is not whether x402 can settle payments. TRM’s dataset already shows it can, at scale in transaction count, and almost entirely in USDC. The deciding question is whether the ecosystem can credibly label which flows are commerce, which are noise, and which are actually initiated by agents rather than scripts.
If registration and reputation become standard inputs to the payment flow, the setup starts to look structural rather than narrative-driven because “agent volume” stops being a vibes metric and becomes something counterparties can price and monitor. Until then, the cleanest read is rails-first: x402 is a functioning USDC settlement path, and “agentic commerce” remains mostly an attribution problem, not a volume problem.
