The x402 Counter Has Shown the Same Four Numbers Since March

Summary: The number the agent payments market runs on comes from a counter on x402.org, under a label that reads "Last 30 Days." It shows 75.41M transactions, $24.24M in volume, 94.06K buyers and 22K sellers.

CoinDesk quoted those same values on July 15, reporting "about 75 million transactions over the past 30 days... moving about $24 million between some 94,000 buyers and 22,000 sellers." The $24 million figure was circulating in March. It was still on the page when I checked it on August 27, on August 31, on September 3 and 4, and again on the morning of September 6. The page carries no last-updated stamp, no chain breakdown and no methodology, and x402.org/faq returns a 404.

Meanwhile the one team that filtered the same underlying data, Artemis working with Visa, cut 178.3 million raw transactions and roughly $135.7 million in raw volume down to 109.6 million transactions and about $15.0 million, as of April 21. The same report counts roughly 5,300 sellers for the entire life of the protocol, against the 22,000 the counter displays.

On August 23 Lincoln Murr, Coinbase's head of AI product, told CoinDesk that 25% to 30% of x402 transactions "may have been generated by users trying to climb public leaderboards rather than buying services they needed."

Thesis: Forty of the largest payment and technology companies on earth, Visa and Mastercard and Amex and Stripe and Google and AWS among them, joined a foundation this summer to standardize how software agents pay each other.

The market they joined has one public number, and that number comes from an unstamped counter that hasn't moved in six months. Every downstream estimate inherits it. The only serious attempt to clean it removed 89% of the dollar volume.

The protocol's commercial sponsor now concedes a quarter to a third of the activity is scoreboard farming. And the explorer best positioned to settle the question sells access to its data, over the protocol it measures, which means the measurement layer of the agent economy is itself paywalled by a participant. Anyone sizing this market is working from a screenshot.

The counter that doesn't roll

A rolling thirty-day window is a promise about arithmetic. Yesterday drops off, today gets added, the number moves.

The counter on x402.org shows 75.41M transactions and $24.24M in volume under the words "Last 30 Days." CoinDesk reported those exact values in mid-July. The same dollar figure was in circulation in March, six months ago, and a16z's own write-up on agent payments attributes a $24 million thirty-day figure to Bloomberg reporting that cites x402.org, which makes the a16z post a characterization of coverage of the counter rather than an independent read. I checked the page directly on August 27, on August 31, twice in the first week of September, and again on the morning this went out. Same four values every time.

What was verified here is the match. Why the numbers match is not established, and I'm not going to guess at it. A page can freeze because a job stopped, because the label is wrong and the figure is cumulative, or because nobody has looked. Any of those explanations produces the same practical result, which is that the most-quoted number in agent payments describes some window nobody can name.

Everyone quoted it anyway

The counter became popular because it was the only thing to quote.

CoinDesk built a July 15 headline around it, and the headline was actually the skeptical read: 75 million payments moving just $24 million. Six weeks later, in the August 23 piece, the same outlet reports "roughly $24 million moved over 30 days in July," which is the frozen value again, now with a month attached to it that the counter never specified. This is how a stale figure acquires a date. Somebody reasonable cites a page, somebody else cites the citation, and by the third hop the number has provenance without having a source.

Nobody involved did anything careless. There is no ongoing public series with a methodology, a timestamp and a chain breakdown. The one report that has all three is a snapshot that was already three months stale on the day it was published, which is a strange thing to be able to say about a market that Visa, Mastercard, American Express, Ripple, Stripe, Shopify, Google, AWS, Cloudflare, Circle and Coinbase all joined a foundation to support.

The x402 Foundation went live under the Linux Foundation on July 14 with 40 members across three tiers. Its market has worse public statistics than a mid-sized DeFi protocol.

The one filtered number removed 89% of the money

Artemis, working with Visa, is the only group that has published a before and after.

Their report, "Agentic Payments from the Ground Up," dated July 2026, takes roughly 178.3 million raw transactions and about $135.7 million in raw dollar volume and reduces them to 109.6 million transactions and roughly $15.0 million.

That's 39% of transactions and 89% of dollar volume classified as wash, test or internal transfer rather than someone paying for something. Average payment after the cut: fourteen cents. Every figure in it is as of April 21, 2026, which the report states plainly at the top of its methodology page and which almost none of the coverage carries forward.

Also from the report, we get a sense of how Artemis is counting the numbers. A buyer is any wallet with at least one adjusted transaction. A seller is a wallet that has received at least three adjusted transactions from at least two different buyers, a threshold the report says is there to strip out self-dealing and test endpoints.

The wash filter is thinner. In the report it gets a single sentence, aimed at transactions that resemble wash trading, whether driven by attempts to inflate protocol metrics or by developer testing.

The tests do have names, just not in the report. a16z's March write-up credits the filter to Lucas Shin, an analyst at Artemis and the report's named contributor, and describes it as flagging wallets that repeatedly transacted with themselves or cycled funds between addresses. What's missing everywhere is a threshold, a lookback window, a cluster definition or any code, which is what anyone would need to reproduce an 89% cut.

The money is shrinking while the transaction count holds

The counter freezing matters more because the underlying series is moving fast in the wrong direction.

Jamie Coutts at Helios Analytics put x402 daily settlement volume down 93% year to date and down 55% over three months, with a seven-day average around $41,800 and a latest daily figure near $28,400, against Q4 2025 peaks that repeatedly approached $800,000 a day and occasionally passed a million.

Token Terminal's read on the week of August 17 counted 8.7 million transfers moving roughly $367,950, against a prior high of 20.1 million transfers moving $10.017 million in the week of November 17, 2025. Transfers fell by half. Dollars fell by 96%. Average transfer size is now about four cents.

The scorekeeper charges admission

Which brings up who actually produces these numbers, because the answer explains why the gap persists.

x402scan is the de facto explorer, and it's operated by Merit Systems, whose own about page says it maintains "the reference indexes for the two leading agent payment standards." Merit is also a General member of the x402 Foundation, which x402scan doesn't disclose on the site. Merit sells services that are themselves paid over x402. And x402scan's data API is x402-paywalled: a request to its merchants endpoint returns HTTP 402, and its OpenAPI spec labels the data endpoints "Micropayment Required." The index charges for access to the index, over the protocol the index measures.

The alternatives are volunteers and they behave better. agenteconomy.to, built by an X handle with no clear identity attached, publishes a four-step methodology, refreshes hourly, and puts this warning on its own stats page: "Raw settlement counts include tests, infrastructure traffic, repeated service calls, and self-directed usage, activity that is real at the protocol layer but not the same thing as verified organic end-user commerce."

Major Matters, the volunteer tracker, is maintained by Charlie Major, Product Development Manager at Mastercard, with a standing disclaimer that the views are his own. His description of why the tracker exists: "The canonical list, because nobody else is keeping one." He also runs Major Labs, an independent research studio working the same market, and the tracker carries its own funding policy: no sponsorship money, no affiliate placements, and a disclosure whenever a tool it covers has a commercial relationship with it.

The honest case against this read

The strongest objection is that this is what a new market looks like sixteen months in, and complaining about its statistics is complaining about weather. Every payment rail was badly measured early. ACH ran for years on numbers the Fed assembled after the fact, and card networks didn't publish anything comparable to a public dashboard until they had regulators asking.

Bad measurement in a young market is an ordinary condition, and a $15 million adjusted market doesn't warrant Visa-grade telemetry.

There's a version where the counter is simply mislabeled, the cumulative figure is roughly right, and the whole thing is a front-end bug rather than a fact about the market. I'd put real weight on that. A stale page is a much smaller story than a fabricated one, and nothing here establishes intent by anyone.

What keeps me on the other side is who's already spending against these numbers. Mastercard shipped Agent Pay for Machines in June with thirty-plus supporters, and its release name-checks x402. Cloudflare announced a stablecoin-settled Monetization Gateway on July 1, though it's waitlist and early access rather than generally available.

Seventeen premier members are staffing this, and somewhere in each of those companies a person is writing a memo that sizes the opportunity.

The asymmetry that bothers me most is the comparison class, and why this feels like an Open Money issue.

x402 settles on public chains, roughly nine transactions in ten on Base, so anyone with a node and a free weekend can count it, argue with the count, and publish something that makes the protocol look bad. That is precisely what keeps happening.

Google's AP2 and UCP are authorization layers rather than settlement rails, so there's no equivalent number to withhold. Stripe's Machine Payments Protocol does settle onchain, on Tempo, and publishes nothing itself; the only public figures for it are the ones Artemis produced in passing, roughly $25,000 across about 115,000 adjusted transactions in its first thirty-three days.

So the rail that can be audited is the one taking the damage, and it's taking it for being auditable. That's a bad thing to teach a market that is still choosing where to settle.

What to watch

Whether the x402.org counter ever moves, and whether anyone at the Foundation puts a timestamp and a methodology on it. Forty members and a Linux Foundation charter is enough institutional weight to fix a stats page. Six months of the same four numbers is a choice at this point, even if it started as an oversight.

Whether x402scan's paywall holds. An explorer owned by a foundation member, charging for the reference index over the protocol it indexes, is a governance problem dressed as a business model. Watch for either a free tier with a methodology, or a competing index.

Strategic implications

If you're sizing this market. Don't put the $24 million figure in a deck. Use the Artemis and Visa adjusted number, roughly $15 million cumulative through April 21, 2026, cite the report rather than coverage of it, say cumulative rather than monthly, and note that the wash filter behind it is unpublished.

If you're building on this rail. The measurement gap is a real product. There's no public series with a stated method, a timestamp, chain coverage and a wash filter anyone can audit, and the demand for one is forty premier and general members deep.

If you're a Foundation member. The protocol's public statistics are worse than its adoption, and it's costing the standard credibility every time a skeptic looks.

If you write policy. The rail with the checkable numbers is taking the reputational damage, and the rails with no numbers aren't. Any reporting regime for agent payments should ask systems that settle privately for the same figures it asks of the ones that settle in public, or the effect is to reward opacity.

Coinbase Tokenized Stocks on Base (2026)

Four US stocks went onto a public blockchain this week in self-custodied wallets, issued by an Abu Dhabi special purpose vehicle whose prospectus reserves the power to freeze and confiscate them.

Daniel McGlynnDaniel McGlynn

Sources

[1] x402.org. Counter observed August 27, August 31, September 3, September 4 and September 6, 2026. Identical values on every observation, with no last-updated stamp on the page. https://x402.org

[2] CoinDesk. "Ripple joins card giants backing x402 as 75 million payments move just $24 million." July 15, 2026. https://www.coindesk.com/tech/2026/07/15/visa-mastercard-and-ripple-join-the-standard-letting-ai-agents-pay-in-stablecoins

[3] CoinDesk. Francisco Rodrigues, "Crypto's next billion users might be AI agents, and they're paying with stablecoins." August 23, 2026. https://www.coindesk.com/business/2026/08/23/crypto-s-next-billion-users-might-be-ai-agents-and-they-re-paying-with-stablecoins

[4] CryptoDaily. Idris Calloway, coverage of the Visa and Artemis agentic payments report. August 20, 2026. https://cryptodaily.co.uk/2026/08/ai-agents-blockchain-activity-token-value

[5] Artemis and Visa. "Agentic Payments from the Ground Up," July 2026. All data as of April 21, 2026. Raw cumulative totals p.3 and n.1; methodology and the adjusted buyer and seller definitions p.5; adjusted x402 totals p.10; buyer concentration and seller count p.12; MPP totals p.13; commissioned-and-funded-by-Visa disclosure p.27. Visa's landing page for the report returned an error on August 27 and August 31, 2026 and was reachable again on September 6. https://www.visa.com/en-us/thought-leadership/innovation/agentic-payments-from-the-ground-up

[6] a16z crypto. Noah Levine, "The honest number behind AI agent payments," March 11, 2026. https://a16zcrypto.com/posts/article/ai-agent-payments-honest-number. Credits the wash filter to Lucas Shin of Artemis Analytics and links his X post. Artemis's own account states the same filter description at https://x.com/artemis/status/2031768320081277029

[7] agenteconomy.to. Methodology and stats warning, observed August 27, 2026, page stamped August 26. https://agenteconomy.to

[8] x402scan, operated by Merit Systems. Merchants API endpoint returns HTTP 402; OpenAPI spec labels data endpoints "Micropayment Required." Observed August 27, 2026. https://x402scan.com

[9] Jamie Coutts, Helios Analytics. Posted to X on August 12, 2026, pointing to a Helios chart of x402 settlement volume from October 2025 through July 2026. https://x.com/Jamie1Coutts/status/2087442066716676368. Figures as reported by CryptoPotato (Wayne Jones, August 12, 2026), https://cryptopotato.com/x402-volume-plunges-93-ytd-as-agentic-ai-economy-hype-fades/, and CCN (Giuseppe Ciccomascolo, August 13, 2026).

[10] Token Terminal figures via The Cryptonomist. Alessia Pannone, "Agentic Stablecoin Payments Reach New Record in 2026," August 25, 2026. https://en.cryptonomist.ch/2026/08/25/agentic-stablecoin-payments-record/. The article reports 8.7 million transfers in the week of August 17, 2026, more than double the prior week's 4.1 million, moving about $367,950, against an all-time high of 20.1 million transfers moving $10.017 million in the week of November 17, 2025, of which $9.78 million moved on Base.