The narrative that artificial intelligence agents are actively spending money across the web has faced a reality check from TRM Labs’ research analyzing 198.9 million transaction settlements worth a combined $52.7 million. All of this financial activity was processed through Coinbase’s x402 protocol spanning Base, Solana, and Polygon networks since May 2025. From this dataset, researchers estimate that only 0.6% to 7.5% of the remaining commerce volume was purely driven by AI agents.
That conservative estimate emerged after TRM filtered gross volume and isolated $25.62 million as likely legitimate commerce. The research team excluded self-payment transactions, bulk flows originating from only one or two payers, and volume directed toward merchants with fewer than ten registered customers. Activity recorded in these net funds late in 2025 included meme token minting traces and payment flows to a single AI analytics service.
Why Transaction Footprints Are Hard to Distinguish
The primary challenge faced by researchers stems from the nature of blockchain data, where standard automated scripts, scheduled jobs, network load tests, and self-dealing leave the exact same on-chain footprint as AI agents. Coinbase launched the x402 protocol in 2025 to integrate payments directly into web requests. The workflow is streamlined: the buyer receives a quote and signs an authorization, while a facilitator verifies its validity and submits the transaction directly to the blockchain.
To identify non-human buyers across this infrastructure, TRM designed screening parameters focusing on wallets with varied payment patterns averaging under $1 over several months. To pass the filter, wallets also had to show agent registration on public domains or proof of transfers to multiple different merchants, rather than simply purchasing a single service repeatedly.
Strict Screening That Could Be Flawed
Researchers also acknowledged that their strict criteria leave an analytical blind spot that might disqualify genuine intelligent bots. A legitimate AI agent specifically tasked with repeatedly purchasing the exact same service would inevitably fail TRM’s test, as its monotonous spending pattern appears on the blockchain ledger just like a routine script.
The study’s conclusion directly challenges widespread market claims regarding the rise of an autonomous agentic economy. While machine-to-machine protocol flows are indeed processing millions of settlement transactions, most parties signing payment authorizations do not yet match the narrative of autonomous entities promoted across the web3 ecosystem. Reported via Decrypt.
Read also: What Is DeFi (Decentralized Finance)?
Read also: Tokenized Stock Market Hits $29.5 Billion - But Your Voting Rights Might Just Be an Illusion
Disclaimer: This article is for informational and educational purposes only, not financial advice. Cryptocurrency assets are highly volatile and carry significant risk. Always do your own research (DYOR) and never invest more than you can afford to lose.
