The emerging infrastructure for autonomous artificial intelligence agents to spend money online may be developing faster than the agents themselves.
A new analysis by blockchain intelligence company TRM Labs found that only a fraction of payments flowing through x402, one of the most prominent protocols designed for machine-to-machine commerce, can be attributed to autonomous AI agents.
TRM examined about $52.7 million in value across 198.9 million x402 settlement transactions on the Base, Solana and Polygon blockchains starting in May 2025, the analysis said. After filtering out activity unlikely to represent genuine commerce, the firm estimated that just 0.6% to 7.5% of the remaining payment value appeared to be agentic.
The findings challenge a common assumption that rising x402 transaction volumes provide a proxy for the growth of autonomous commerce. Ordinary software scripts, scheduled processes and even self-dealing can produce blockchain records indistinguishable from payments initiated by AI agents, Decrypt reported Sunday (Sept. 13) in its summary of the research.
Launched by Coinbase in 2025, x402 effectively turns the long-dormant HTTP “402 Payment Required” status code into an internet-native payment mechanism. When a buyer or bot requests an online resource, such as an API call, the server responds with a price. The buyer then resubmits the request with a signed payment authorization. A facilitator verifies the authorization, broadcasts the transaction to a blockchain and pays the network fee. Most such services operate on a pay-per-call basis.
That structure removes the conventional checkout page and makes x402 well suited to autonomous agents. But critically, the protocol does not require AI. A conventional script can execute exactly the same sequence.
To distinguish commerce from other activity, TRM removed addresses paying themselves, large flows dominated by one or two payers, and sellers with fewer than 10 distinct buyers. That reduced the original $52.7 million to $25.62 million of likely commerce, per the analysis.
TRM then applied tests for identifying agents. One model looked for facilitator-broadcast payments with varying amounts averaging less than $1. Another looked for sustained activity over multiple months and evidence such as registration in a public agent registry or payments to multiple sellers, the analysis said.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
The methodology reflects a limitation in measuring agentic commerce. Agency itself is not visible on-chain. TRM assumed an autonomous agent would be more likely to explore multiple products and services, while an address repeatedly paying the same amount to the same service resembles conventional automation, according to the analysis.
But TRM acknowledged that assumption could undercount genuine agents. A single-purpose agent repeatedly purchasing one service, for example, could look exactly like a script, the analysis said.
The composition of x402 activity is also evolving rapidly. Late-2025 traffic included activity consistent with meme-token minting and payments to an AI-analysis service. Volume subsequently became concentrated in a single payment contract before AI and agent services reappeared through an agent-payment router around mid-2026, per the analysis.
One feature is already consistent, however. The analysis said that x402 is a stablecoin payment system. TRM found in the analysis that $52.47 million of $52.68 million in settled value, or 99.6%, used USDC. Nearly every merchant identified by TRM specified USDC as its preferred payment asset.
The findings also point toward a developing compliance problem. Existing agent registries allow people to declare ownership of agent addresses, but participation is voluntary and uncommon. Meanwhile, agentic payments can combine very low dollar values with extremely high transaction counts, potentially defeating controls designed around human payment patterns, the analysis said.
The infrastructure therefore needs better agent registration, machine-readable counterparty reputation and monitoring systems designed around transaction volume rather than simply dollar thresholds, according to the analysis.
“Agentic commerce will need agentic compliance,” TRM said in the analysis.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.
Add as Preferred Source
