Artificial intelligence agents are no longer experimental tools — they are active economic participants, as evidenced by new statistics from Base. In the 30 days leading up to May 29, 2026, AI agents completed 3.1 million onchain transactions and transferred $1.2 million using wallets and stablecoins on the Ethereum Layer 2 network. These agents purchased a wide range of services, including research data, financial market tools, travel bookings, and developer infrastructure, often without requiring constant human approval.
The shift represents a significant expansion from the earlier wave of AI agent activity. Between October 2024 and February 2025, nearly 16,000 agents launched on Virtuals.io, a platform for creating AI agents and tokens, but their use was largely limited to social interactions and community engagement. Since then, improvements in AI models — enabling planning, coding, research, and execution — combined with easier wallet integration, increased stablecoin usage, and ultra-low transaction costs, have turned agents into direct customers. Service providers like Venice (AI inference), BlockRunAI (access to 50+ AI models with USDC settlement), Exa (search), and Wolfram Alpha (computation) now offer pay-per-use access. In the travel sector, Tripadvisor, FlightAware, and Amadeus are enabling autonomous bookings. Infrastructure giants are also involved: Cloudflare supports x402 payments, while Amazon Bedrock AgentCore Payments integrates x402 and Coinbase wallet infrastructure for enterprise workflows.
Base and its parent company Coinbase are actively building infrastructure for this agentic economy. Coinbase CEO Brian Armstrong urged businesses to “make sure your business is ready to accept AI agents as customers.” The company stated that future developments will focus on “earning agents” that generate income — examples include Felix, which reported over $261,000 in revenue, and Kelly Claude, which sells apps and digital products. Meanwhile, the broader industry is embracing autonomous finance. Robinhood launched Agentic Trading and Agentic Credit Card services with user-defined limits, while Coinbase introduced Base MCP, a system connecting AI assistants like ChatGPT and Claude to crypto wallets for swaps and portfolio monitoring. Coinbase’s x402 payment infrastructure, designed for machine-to-machine transactions, has gained significant traction: AI agent transactions use USDC 99% of the time, and over 90% occur on Base, with monthly volumes surpassing 75 million transactions.
Decentralized finance is emerging as a key testing ground. Andrew Isaacs, COO of Neyro, a decentralized AI trading platform, argued that DeFi’s fast-moving, non-custodial nature makes it an honest environment for evaluating AI reliability. “In trading, it can show whether this new model of automation is actually reliable under pressure,” he said. However, he cautioned against sacrificing decentralization for convenience. Security concerns remain paramount: when Base launched MCP, it stressed that transactions require explicit user approval and never access private keys; researchers from Google, Meta, and others have urged treating AI agents as untrusted components. Despite these challenges, Isaacs sees a clear use case: “A human trader cannot watch every pool, every token. But an AI agent can.”
