The infrastructure for autonomous AI agents to spend money online may be developing faster than the agents themselves. That is the core finding reported by PYMNTS: agentic payments are growing, but most payments flowing through x402 — the HTTP-native stablecoin payment protocol pushed by Coinbase and Circle — are not initiated by AI agents at all. We wrote last week that x402’s claimed 100 million agentic payments deserved a harder look; this reporting adds a data point that sharpens the same skepticism.
What the numbers actually show
According to the PYMNTS report, transaction volume through x402 is expanding, but the majority of payments on the protocol do not originate from autonomous agents. The protocol has effectively become a general-purpose, stablecoin-based pay-per-request layer for HTTP APIs, and humans or conventional software clients are using it accordingly. This is not necessarily a failure — it is a different adoption curve than the machine-customer narrative implies. The distinction matters for anyone sizing the agentic commerce market: protocol volume is not the same thing as agent volume.
Why infrastructure runs ahead of agents
Building payment rails is cheaper and faster than building agents reliable enough to be trusted with wallets. x402’s design — a 402 Payment Required HTTP response, an ephemeral keypair per payment, and USDC settlement — solves the machine-payable API problem cleanly. But the demand side requires autonomous agents that businesses are willing to let spend money unsupervised, and that bar is higher. The result is a familiar pattern from payments history: rails get laid years before the killer application arrives. Card networks, and before them wire rails, followed the same sequence.
How to read adoption metrics going forward
The practical lesson for analysts is to separate three numbers that get conflated: protocol transaction counts, payments initiated by software rather than humans, and payments initiated by genuinely autonomous agents. Only the third supports the agentic commerce thesis. The recent skepticism around x402’s headline volume figures — including on-chain analysis questioning how much activity is genuinely agent-originated — suggests issuers and developers will need to publish better attribution before the market prices this in. Standards work like BNB Chain’s agent lifecycle draft and Ant International’s protocol face the same measurement problem.
What to watch
Watch for issuers and analytics firms publishing agent-attribution breakdowns rather than raw counts. If agent-initiated share of x402 volume climbs steadily, the machine-customer story strengthens. If it stays flat while total volume grows, x402 has still succeeded — just as an HTTP-native micropayment layer, which is a smaller and less novel claim. Both outcomes are viable; only one of them is the thesis being sold.