Research · Part I
What x402 services actually do when you try to pay them.
Public explorers can tell very different stories about the x402 economy — the emerging world where software agents pay for APIs in stablecoins, no human in the loop. One view emphasizes growth: many transactions, volume climbing. Another is far more skeptical about how much is real. They’re often indexing the same services, which makes it hard to know, from the outside, what’s actually happening.
We wanted to measure it ourselves — evidence, not assertion. If an agent is about to pay one of these services, is the thing actually alive, honestly priced, and really used? So for two weeks we knocked on every listed x402 service several times a day, read what it says back, watched the money on-chain, and committed the whole record to a public, timestamped log. Here’s what the machine-payment economy looks like when you actually try to pay it.
Every number here comes from what services and the chain themselves report. We record the payment challenge each endpoint returns, pair it with on-chain USDC receipts — who actually paid whom — and anchor every sweep to Bitcoin, so the history can’t be quietly edited after the fact. All of it is reproducible from the public record. Evidence, not opinion.
Start with scale. The directories list roughly 8,000 x402 endpoints. Collapse the duplicates and they resolve to about 1,000 distinct services. Of those, 569 return a valid, priced payment challenge when you knock. The rest — around 40% — answer with a 404 or the wrong method, sit dead, or reply in a way no standard client can parse.
Four in ten “machine-payable services” can’t be confirmed alive and honestly priced at any given moment. That’s the verification gap, and it’s the single most useful thing to know before you point an agent at a registry.
The good news: when a service does answer, its live price usually matches what it advertised. For about 94% of priced services, listed and live agree within rounding. Price dishonesty is not rampant.
The tail is where it gets sharp. Fourteen services quote at least double their listed price once you actually ask, and the worst charges twenty-three times more — a service listed at $0.078 that demands $1.85 at the door. (An earlier sweep caught one listed at $1 that demanded $10,000; it has since vanished.) On rails with no chargebacks, a surprise like that is money gone, silently. Rare — but exactly the thing you want caught before your agent pays, not after.
Prices are what a service claims. On-chain receipts are what actually happened. So we read the USDC flowing into the busiest pay-to wallets, and the answer is more interesting than either the hype or the doom.
Concentration is not wash. The wallets collecting for the most services turn out to be, mostly, real businesses — a median of 29 distinct buyers each and a 56% repeat-buyer rate. More than half of customers come back. One wallet has served 183 distinct buyers across 8,698 payments at an 88.5% repeat rate. That’s not padding; that’s a product people return to.
But real demand is small. Add up the lifetime USDC across the forty biggest earners and you get about $42,000. The median busy service has moved twenty dollars, total. This is a genuine micropayment economy — and it is still tiny.
And wash exists, but it’s the minority. Roughly one in seven of the concentrated wallets shows the tell-tale pattern: a single wallet cycling most of the volume, or an address paying itself. Present, nameable, not dominant.
The honest picture is the unglamorous middle: real paying demand that even retains, at small scale, with a measurable minority of fakery. Not millions of transactions. Not all wash. Somewhere true, in between.
One more thing, because it’s the whole point of measuring in the open. A week earlier our own numbers said ~280 services were “broken.” They weren’t. They deliver the standard payment challenge in an HTTP header instead of the response body — and we were only reading the body. We fixed the parser; the next sweep recovered them wholesale. A large slice of what we’d have reported as “broken” was our own blind spot.
We’d rather find that ourselves, in public, than have you find it for us. A measurement is worth exactly as much as its willingness to be wrong out loud.
Everything above is measured. This part is a bet — and we’ll flag it as such.
Right now, “which service should my agent trust?” isn’t a question many people are asking, because most of the effort still goes into making a single payment work at all. Fair enough; the space is early. But look at how human commerce handles money moving between strangers: audit trails, know-your-counterparty, independent ratings, the ability to prove you did your diligence. None of it stays optional once the value at stake grows, once someone is liable when the wrong party gets paid, and once the rules catch up. There’s little reason to think autonomous agents moving real money will be the exception.
And the rules being unwritten is not a licence to build carelessly — if anything, the opposite. When an agent pays on your behalf, you are accountable for what it did, regulation or no regulation. Accountability needs something to point to: a way to show that the counterparty was real and verified at the moment of payment, that anyone can check for themselves. Operating safely and answerably shouldn’t wait for someone to make it mandatory.
None of that framework is written yet, and we’re early — deliberately. But the thing that becomes valuable when it arrives isn’t a verdict you take on faith; it’s an independent, continuous, tamper-evident record of who did what, that anyone can recompute. That’s the layer we’ve been building and anchoring since day one — so that as agents begin to choose and pay for services on their own, there’s a way to steer them toward the safe ones, and a way for the people behind them to stand behind what they did.
Verantis is a neutral measurement layer for machine payments. Every sweep is public and Bitcoin-anchored, and the underlying data, the directory, and an agent-queryable interface are open for you to check and poke.
Numbers in this piece are from the Verantis sweep of 2026-07-25 and a lifetime on-chain scan of the highest-volume pay-to wallets. Methodology and raw records are public and reproducible. Figures describe the ecosystem at that date; the live directory reflects current numbers.
This report presents independent measurements and our own analysis of them. Measured figures are point-in-time observations and may change; interpretations and any forward-looking views are our own and may differ from others’ assessment of the market. Nothing here names or alleges wrongdoing by any specific service, and nothing here is financial, investment, or legal advice.