Coinbase is taking a significant step toward integrating artificial intelligence into mainstream finance. CEO Brian Armstrong revealed on September 9 that the exchange is building a financial account designed specifically for AI agents, building on existing tools that already allow software to trade within controlled portfolios. The announcement came in response to a question from Ruby on Rails creator David Heinemeier Hansson about the possibility of an “agentic bank” for autonomous routine spending. Armstrong did not provide details on launch dates, product names, fee structures, or regulatory frameworks.
Coinbase for Agents: Controlled Trading Today
Coinbase for Agents currently enables supported AI applications to connect with Coinbase Advanced Trade through a remote Model Context Protocol (MCP) server or a local command-line interface. This setup gives agents access to designated portfolios rather than a customer’s entire account, ensuring a layer of security and control.
Coinbase advises users to create a separate portfolio and fund it only with assets they are willing to expose. Customers can also limit an agent’s permissions to that portfolio, reducing the funds available for unexpected orders. The service supports spot trading across more than 900 cryptocurrency pairs. Eligible US customers can also access futures, S&P 500 equities, portfolio monitoring, and USDC-to-dollar conversions, subject to regulatory and account restrictions.
Portfolio Limits Define the Current AI Model
Agents can transfer assets between authorized Coinbase portfolios through approved API permissions, but those permissions do not allow withdrawals to external blockchain addresses. Customers can revoke an application’s access at any time through Coinbase security settings. Models such as Claude or ChatGPT can connect through a local MCP server, operating through an isolated sub-account with user-defined transaction limits rather than unrestricted access to cards or outside wallets.
The question now is whether Coinbase will extend these controls from trading into routine payments and merchant spending. The company has not disclosed whether the planned AI financial account will support cards, bank transfers, recurring bills, refunds, or direct merchant payments.
x402: A Payment Protocol for AI Agents
Coinbase is also developing x402, a payment protocol designed around pay-per-request transactions. This model lets AI software pay for individual actions from a USDC balance without relying on conventional subscription flows. Currently, x402 works through Agentic Wallet rather than direct integration with Coinbase for Agents. Direct x402 support inside Coinbase for Agents represents another planned step alongside any broader financial account announcement.
Broader AI Integration and Strategic Moves
The company has expanded its use of AI internally and across customer products. Earlier in 2026, Armstrong cut about 14% of Coinbase’s workforce, or roughly 700 employees, while shifting toward smaller AI-native teams. In August, AI agents gained access to derivatives and S&P 500 stocks. Coinbase also tested autonomous travel planning through AWS AgentCore, including USDC hotel payments on Travala using Base.
Meanwhile, Coinbase continued expanding beyond spot cryptocurrency trading. The exchange filed two SEC registrations for US stock perpetuals, although those filings did not establish a product launch date. The company also appointed Anthony Armstrong to its board in early September.
As Coinbase moves forward, it has not detailed how identity checks, disputes, refunds, or legal responsibility would work when software initiates transactions. These questions remain central to the future of AI-driven finance.
Leave a Reply