All posts
· 16 min read ·

Most agent payments are x402 micropayments. Most of the dollars aren't.

Of 150 random Base transactions into indexed agents' payment wallets, 143 were settled by a known x402 facilitator. Of the 150 largest, none were, and over 30 days facilitator-settled transfers carried 4.8% of $874,561 in classified inflow. What settlement-path attribution shows, and what it can't.

Count the transactions arriving at the payment wallets of indexed agents and the agent economy looks like x402. Of 150 Base ledger transactions drawn at random, 143 were settled by a known x402 facilitator, nearly all of them by Coinbase. Count the dollars and it looks like something else. Of the 150 largest transactions in the same ledger, none were. And across the 30 days ending October 5, transfers settled by a known facilitator carried 4.8% of the $874,561 of inflow we have classified. Three-quarters arrived as plain token transfers sent by the payer.

Both lenses describe the same wallets. They describe different payment activity. Most transactions are small x402 settlements. Most of the money is ordinary on-chain payment that happens to land in an address an agent also publishes as its payTo. Any revenue figure for "the agent economy" that sums inflow into declared wallets, including the one Agenstry has published for months, is mostly measuring the second population.

What a Transfer log can't tell you

Agenstry's observed inflow has always been a deliberately narrow number: stablecoin transfers into wallets that agents declare on their own cards, counted once per wallet, with an explicit caveat that it is not verified sales or x402 revenue. In August we measured one way that number overstates. Summing per agent instead of per wallet counted 17.41% of the same money twice, because 71.9% of earning agents shared a settlement wallet with another listing.

The second gap is in the data source itself. An ERC-20 Transfer event records who paid whom and how much. It does not record how the transfer was submitted, and the submission path is where x402 leaves its mark.

In the x402 exact scheme on EVM chains, the buyer signs an EIP-3009 authorization, and a facilitator submits it on chain by calling transferWithAuthorization on the token contract. EIP-3009 was written for exactly this: it lets a token holder delegate gas payment to someone else. Coinbase's documentation describes its facilitator as the party that "validates signed payments, screens transactions, submits settlement onchain", and states that with exact, each accepted payment settles in its own on-chain transaction. The facilitator's signer is therefore the transaction's sender, even though the token moves from the buyer to the seller.

x402 also has a self-describing marker. Its builder-code extension appends an ERC-8021 suffix to the settlement calldata, carrying an a code for the app that exposed the endpoint, an s code for client attribution and a w code for "the facilitator that settled the payment on-chain". The facilitator writes that suffix at settlement time. It uses the same suffix format as Base's builder codes, which off-chain indexers extract after the fact.

None of this appears in the Transfer log. All of it appears in the transaction.

The measurement

On October 5 we started reading each ledger transaction once more with eth_getTransactionByHash and placing it in one of four groups. The method is on our methodology page:

  • facilitator: the transaction was sent by a known x402 facilitator signer. The signer list is vendored from the facilitators package in Merit Systems' x402scan, plus PayAI's own published signers, which gives 32 facilitators, 128 EVM and 27 Solana addresses as of October 5.
  • direct_transfer: the payer itself called transfer on the token.
  • contract_call: everything else, including routers, swaps, smart wallets, and EIP-3009 authorizations relayed by an address that is not on the list.
  • unknown: the node returned no transaction.

Every transfer of $1 or more is classified, plus a fixed 1-in-64 sample of smaller ones, chosen by transaction hash. Shares are reported against classified value, with the classified fraction printed beside them, so "not yet read" is never silently treated as "not facilitator-settled". The surfaces say "settled by a known x402 facilitator" and never "is x402". A facilitator can settle anything, and a facilitator missing from the list reads as a contract call.

Same wallets, two lenses: how inflow into agent payment wallets was settled Three 100-percent bars for Base ledger transactions into indexed agents' declared payment wallets. 150 random transactions: 143 settled by a known x402 facilitator, 7 other. 150 largest transactions: none facilitator-settled, 134 plain transfers sent by the payer, 16 routers or smart wallets. 30-day classified value of 874,561 dollars: 4.8 percent facilitator-settled, 75.6 percent plain transfers, 19.6 percent other contract calls. Same wallets, two lenses How inflow into indexed agents' declared payment wallets was settled on Base By transaction: 150 drawn at random 143 facilitator-settled 7 other By transaction: the 150 largest 0 facilitator-settled · 134 plain transfers 16 routers / smart wallets By value: 30 days, $874,561 classified 75.6% plain transfers 19.6% other 4.8% facilitator-settled Pink: settled by a known x402 facilitator. Violet: plain transfer sent by the payer. Grey: other contract calls. Rows 1-2: agenstry.com/methodology#settlement, early October 2026. Row 3: agenstry.com/flows, snapshot 2026-10-05.

Two populations in one ledger

The 30-day split on /flows, snapshot October 5, 2026, with 97.9% of the $893,638 ledger inflow classified:

Settlement path Value Share of classified value
Settled by a known x402 facilitator $41,773 4.8%
Plain transfer sent by the payer $661,088 75.6%
Routers, swaps, smart wallets, unlisted relayers $171,700 19.6%

Facilitator-settled value is concentrated in one operator. Coinbase accounts for $41,024 of the $41,773, or 98.2%, followed by PayAI at $639. The 26,312 facilitator-settled transfers behind that figure average under $2 each, even though transfers below $1 are only sampled. That fits the random sample: at the transaction level, x402 is the common case.

The earnings leaderboard now shows the x402-settled share beside every wallet's inflow, and the 25 largest wallets on /flows separate cleanly into two shapes. Several wallets are 100% facilitator-settled with very small average payments. One staging deployment received 70,295 transfers totalling $3,739 in 30 days, about five cents each. Another endpoint received 51,521 transfers totalling $1,313, under three cents each. dTelecom's speech-to-text endpoint is the largest wallet in this shape: $32,113 settled by Coinbase's facilitator across 11,402 transfers.

The other shape is large and rare. The largest wallet in the ledger belongs to api.bitrefill.com: $633,340 over 4,890 transfers in 30 days, about $130 each. Its page shows $595.80 settled by a known facilitator, $602,383 in plain transfers sent by the payer and $8,396 in other contract calls. That one wallet is 70% of all classified value in the ledger and 91% of all plain-transfer dollars.

That raises the obvious objection: is the finding just one merchant? Removing Bitrefill's wallet entirely leaves $263,186 of classified value, of which 15.6% was settled by a known facilitator. The share is higher, and still a minority. Most of the remainder, 62%, is contract calls. The leaderboard shows what that looks like at the wallet level: one endpoint took $24,779 in six transfers, another $18,048 in 46, each with $0.00 settled by a known facilitator.

What the split does and doesn't say

Settlement path describes how a payment was submitted, not who sent it. It is tempting to read "plain transfer" as "human" and "facilitator" as "agent". The data does not support either reading. Bitrefill's agent integration documentation on Coinbase's developer platform names the same payTo address our ledger tracks and routes agent purchases through its x402 invoice/pay route. A script can pay through a facilitator, and a person can pay through one too. The classification answers one question: did a known x402 facilitator submit the transaction? Who controlled the paying key is the payer-independence question, which remains open.

A facilitator listing is not settlement evidence. Nansen's API appears in two facilitators' x402 discovery feeds: 45 endpoints in PayAI's and 31 in Coinbase's. Over the last 30 days its wallet shows $1.36 settled by a known facilitator, from 64 transfers, and $7,162 in other contract calls. That does not show that Nansen's customers are not paying via x402. It shows that the money we observed arrived by a path our signer list does not recognise as a facilitator. A discovery feed records what a seller declared. The transaction records how money moved. They are separate evidence and should be reported separately.

Sellers can submit their own authorizations. Bitrefill's $8,396 of "other contract calls" includes transferWithAuthorization calls submitted by Bitrefill's receiving wallet itself, like this $28.70 settlement on September 20. EIP-3009 explicitly allows the recipient to submit the authorization. x402scan lists a Bitrefill facilitator, but under a different signer address, so these x402-shaped transactions fall outside the x402-settled column. The facilitator share is a lower bound on x402 settlement, and how loose that bound is depends on how current the signer list is.

Headline figures need their x402-settled share beside them. "Agents received $X" invites the reading "agents earned $X from x402 calls." On our own surfaces, every inflow figure on the flows page, the leaderboard and agent pages now carries its x402-settled share. We'd suggest the same of any dashboard that reports agent revenue from wallet inflow: state the settlement path, or state that you didn't measure it.

What we're watching

Batching will collapse the transaction lens. x402's batch-settlement scheme replaces per-request transfers with commitments settled later, for example as one contract call redeeming accumulated vouchers. The Solana binding's implementation landed in the x402 repository on September 25, and PayAI opened a Solana public preview on September 30, describing how a million $0.001 payments can settle on chain as one $1,000 claim. Circle's Gateway nanopayments work similarly: Gateway "collects authorizations and settles net positions in bulk onchain". A transfer scan sees one claim where there were a million calls, and that claim reads as a contract call. As batching spreads, transaction counts will understate x402 activity, and the dollar lens will be the only one that still works.

The signer list will drift. Circle's x402 Facilitator Service launched in mid-September for USDC on Arc, Base and Polygon PoS, reported by Crypto Briefing on September 16. Circle broadcasts every transfer and pays settlement gas, but its documentation does not publish signer addresses. Until those appear on a list we can vendor, Circle-settled payments into indexed wallets will read as contract calls. Every new facilitator creates the same gap.

Builder codes may replace sender lists. Coinbase-settled transactions in our ledger carry a builder-code suffix naming its facilitator (w=cdp_facil1, the example on our methodology page). The x402 repository began validating app attribution in builder codes on August 31. If every facilitator writes a w code, attribution can come from the calldata rather than from a list of addresses someone has to maintain. That would make this classification less dependent on us.

Limitations

  • Base dominates. The settlement classification runs on EVM chains via transaction lookups, with Solana handled through the fee payer. In practice, $908,239 of the $910,736 gross 30-day inflow on /flows is on Base, so these results are Base results.
  • The facilitator share is a lower bound. It covers only signers on a vendored list. Unlisted facilitators, sellers settling their own authorizations and batched claims all read as contract calls.
  • Sampling below $1. Transfers under $1 are classified on a 1-in-64 sample, so transfer counts by path are not complete counts. The value shares are robust to this, because value is concentrated in transfers above $1, all of which are classified.
  • Inflow is not revenue. These are gross stablecoin transfers into declared wallets. They are not netted or audited, and they do not prove that a sale or paid call happened. Shared wallets can receive payments for services other than the listed agent.
  • The random sample is small. The 143 of 150 figure comes from a single draw from the early-October ledger. The 30-day value split is the population figure, and the one to cite.

The live figures are on /flows, the earnings leaderboard and every agent page with observed inflow, and the method is at /methodology#settlement. If a facilitator's signer is missing from our list, tell us. The fix is one line in the list, and it moves money between columns in a way anyone can check on chain.

Sources

← Back to blog Agenstry