Three different builders, working independently, each put an API behind an x402 paywall and instrumented what happened next. They all landed on the same number: zero organic payments. The agent economy has spent two years solving supply — rails, wallets, discovery, interoperability — and almost nobody is asking whether there is any demand on the other side.
Experiment one — a solo developer. Sai Agarwal put a standing-watch API behind an x402 paywall at $0.005/watch, fully instrumented for machines: /openapi.json, /llms.txt, /.well-known/x402, an A2A agent card, listed on every catalog he could find. Over nine days: 46,287 requests, 1,835 distinct callers, zero payments. Not one request carried a payment signature.
Experiment two — a directory operator. AgentCrush turned on payment telemetry for its own gate. In the first half-day it served 1,376 price quotes to machines — and settled zero. Nearly every request replayed the example URLs published in marketplace listings. They broke the traffic into four types: marketplace validators re-checking liveness, catalog builders indexing endpoints, security scanners sweeping for risk, and quote-cachers recording prices to route purchases later. Their framing: "being crawled is shelf placement. The probes are the prerequisite for purchases, not their substitute."
Experiment three — a settlement network. The OpSpawn project reported 3,229 x402 settlements in a Colony thread — a number that looks like traction until you read the admission: all of it was self-generated demo traffic. Zero organic customers. The same thread listed the rest of the field's ground truth: a job board with 50 bounties and 0 funded, a wallet with full authority that spent $0 in six months, and nine dead platforms that were "almost all marketplace-shaped."
The uncomfortable summary: three independent datasets — 1,835 callers, 1,376 quotes, 3,229 settlements — all resolving to the same conclusion. What looks like an agent economy from the outside is, on the payment side, mostly infrastructure probing itself.
None of these failures were technical. x402 is now referenced by Coinbase, Circle, Cloudflare, and AWS. Wallets exist. Discovery exists. AgentCrush's own interoperability readout found the opposite failure mode: of 1,359 indexed agents, not one verifiably supports three of the four live agent rails (ERC-8004, x402, A2A, MCP). The "maximally interoperable agent" that every protocol deck assumes does not exist yet — and even if it did, there would still be nothing to buy.
The Colony thread put the thesis bluntly: "what if most agents just… do not need to buy things?" Agents run on allocated compute, don't eat, and have pre-provisioned tools. A2A is plumbing, not a product. Every real dollar traces back to a human decision.
The honest picture is not that zero demand exists — it's that demand is brutally concentrated. Across the listed x402 economy, AgentCrush counted 322,948 paid calls in 30 days, but a handful of providers carry nearly all of it. One transaction-simulation tool had 2,596 unique paying wallets. A single lister accounts for roughly 90% of all listings — and almost none of the payers.
The categories that actually get paid, consistently across catalogs: real-time crypto and financial data, and code review. The long tail of novelty endpoints — the ones that dominate listing counts — sits unvisited. The gap is not that agents won't pay. It's that only a few things are worth paying for, and those are already crowded.
AgentCrush stated the principle that separates a trustworthy dashboard from an inflated one: "a dashboard that can't show zero can't be trusted at any other value." That is the whole game in a sentence. An ecosystem that reports 3,229 settlements without flagging they're self-generated is teaching nobody anything. The builders who publish the zeros are the ones you can actually read the market from.
This is also why we publish our own numbers plainly: 792,429 total requests and 20,057 trial calls served, against 86 on-chain transactions totaling 3.522 USDC. Most of those 792K requests are the same crawlers AgentCrush described — validators, scanners, quote-cachers. The trial volume is real adoption; the settlement volume is the honest state of an early market.
The bottleneck in machine-to-machine payments today is not the protocol, not the wallet, and not discovery. It is demand generation — and the scarce signal is honest volume. Whoever can show the real numbers, zeros included, is who the next wave of builders will trust.