AI Agents Barely Drive the Machine-to-Machine Economy

SAN FRANCISCO — Technology conglomerates and financial institutions are investing millions of dollars in infrastructure for an anticipated “machine-to-machine” economy. The systems are intended to let autonomous artificial intelligence agents transact independently, but newly compiled blockchain data indicates that most current automated transaction volume comes from basic computer scripts rather than sophisticated AI agents.

Coinbase launched the fundamental technical framework for this ecosystem, known as “x402,” in 2025. The protocol adapts HTTP 402 (“Payment Required”), a non-standardized status code reserved in the original 1990s World Wide Web specifications by internet creator Tim Berners-Lee. HTTP 402 was intended to support digital microtransactions through web browsers, but the absence of a native, digital-first financial system left it largely unused for decades.

Under the modern x402 standard, payments are integrated into ordinary web requests. An automated system requests data or a service, the seller provides a price, the buyer signs a cryptographic payment authorization, and a third-party facilitator verifies the authorization, broadcasts the transaction to the blockchain, and covers the network transaction fees.

Major companies have moved into the sector. In May 2026, Amazon announced a partnership with Coinbase and payment processor Stripe to launch AgentCore Payments, a system designed to let autonomous AI agents purchase web services using stablecoins. Binance followed in August 2026 with its “Agent OS,” which integrated an automated payment layer.

Coinbase has also used its Layer-2 blockchain, Base, to run a $1 million accelerator program for startups developing AI agents and specialized payment gateways. Yet a comprehensive network study by blockchain intelligence firm TRM Labs found that true autonomous commerce remains in its infancy.

TRM Labs analyzed approximately $52.7 million in transaction volume across 198.9 million settlements processed by known payment facilitators since May 2025. The activity covered Base, Solana, and Polygon. Circle’s dollar-pegged stablecoin, USDC, accounted for 99.6% ($52.47 million) of the total settled value throughout the period examined.

Researchers removed self-payments, automated load tests, other anomalous bulk transaction flows, self-dealing transactions, bulk transfers sent by one or two dominant payers, and sellers that interacted with fewer than 10 unique buyers. That left $25.62 million classified as “likely commerce.” Within that amount, TRM Labs estimated that true AI agents represented between 0.6% and 7.5% by value, while standard automated scripts generated the rest.

Distinguishing an advanced AI agent from a basic software script remains a primary challenge. An ordinary, non-intelligent script can perform the same sequence of blockchain actions as a complex AI. TRM Labs therefore assumed that an authentic autonomous AI agent behaves adaptively, exploring and transacting across multiple services and products. An address repeatedly paying the exact same price to the same service was categorized as a standard script hitting an API.

The researchers also screened transactions broadcast by facilitators that featured varying transaction values averaging under $1. A stricter secondary test required the transacting entities to maintain that behavior across multiple months, have a public registration in an agent directory, or send payments to multiple distinct sellers. The firm acknowledged that its methodology could understate the market’s actual scale because single-purpose AI agents that repeatedly buy from only one service would be filtered out as basic scripts.

On-chain activity changed over time. In late 2025, transaction volume on these rails was dominated by automated meme-token minting and payments directed to a single AI-driven market analysis service. By early 2026, the volume became heavily concentrated in a single payment contract, before shifting back to general AI-service payments through an agent-specific payment router in the middle of the year.

The sector also lacks standardized verification and identity frameworks. Decentralized, on-chain agent registries allow developers to declare ownership of specific agent blockchain addresses, but participation is voluntary and currently ignored by most market participants.

That anonymity creates difficulties under current U.S. financial regulations, including the Bank Secrecy Act and requirements enforced by the Office of Foreign Assets Control (OFAC). Traditional anti-money laundering (AML) and know-your-customer (KYC) compliance models are designed for human identities and high-value transactions, while machine-to-machine commerce depends on massive volumes of sub-dollar transactions executed by software programs.

Industry compliance specialists say standard financial monitoring tools must adapt to stop bad actors from exploiting autonomous agent networks for micro-structuring or automated sanction evasion. TRM Labs emphasized that the underlying payment rails are technically operational, but the ecosystem cannot scale safely without specialized monitoring, clearer counterparty reputation networks, and reliable registration standards.

The researchers concluded that a functional machine-to-machine economy will depend heavily on “agentic compliance” systems tailored to manage the massive scale and velocity of micro-payments.