Coinbase Launches 'Coinbase for Agents,' Letting AI Execute Crypto Trades and Pay for Research Without Human Input

Since Coinbase unveiled AgentKit for developers in 2024 and added an in-app AI assistant last December, the company has been building toward a single goal: letting software agents handle financial transactions without a human clicking through each step. That goal took a major step forward Thursday with the launch of Coinbase for Agents.

What the product actually does

According to TechCrunch and CNBC, users can link the agent to their main Coinbase account or, if they prefer tighter controls, sandbox it in a separate environment. From there, the agent can execute trades on Coinbase Advanced, the company's professional-grade platform, which includes TradingView charting tools.

Users can prompt it in plain language: rebalance a portfolio, follow a stated investment thesis, or execute a one-time trade. As of June 11, the agent supports crypto spot markets and derivatives. Support for equities and prediction markets is planned but not yet live.

The agent also plugs into Coinbase's x402 protocol, an open machine-to-machine payments standard the company developed last year in collaboration with AWS, Anthropic, Circle, and Near. Using x402, the agent can pay for paywalled research APIs and on-demand compute resources without requiring a login or subscription, then trade on the insights it purchases. Settlement runs through USDC on Base, Coinbase's in-house Layer 2 blockchain.

Lincoln Murr, Coinbase's head of AI product, told CNBC: "The whole idea is to give agents access to money and, through that financial independence, improve their set of capabilities to pretty much anything on the internet. In the 2010s, every internet company dealt with the transition from desktop and web into a mobile environment. And now in the late 2020s, we're seeing the exact same thing happen where agents are going to be the new primary economic actors on the internet."

The tool works inside ChatGPT and Claude through a Model Context Protocol (MCP) server, according to TechCrunch.

How Coinbase makes money on this

CNBC was the most useful source here for the business mechanics. Coinbase earns trading fees on every agent-executed trade. On the payments side, it captures fees and spreads on USDC movement used as settlement currency. It also benefits from increased transaction volume on Base. Every agentic transaction is a revenue event for Coinbase across multiple layers simultaneously.

The competitive context

Coinbase is not alone. Robinhood launched its own agent-trading tools days before this announcement, according to TechCrunch. Visa invested in Replit to support agentic payments for developers last month. And Visa struck a deal with OpenAI this week on agentic payment capabilities. The race to own the financial layer of AI-agent activity is moving fast.

The legitimate concern

Giving an AI agent autonomous access to a financial account, including the ability to execute trades in derivatives markets and spend money on external services, introduces compounding failure modes. A misunderstood instruction, a prompt injection attack targeting the agent, or an API outage mid-trade could produce real financial losses before a human even knows something went wrong. Coinbase says it will "soon" add custom limits on trade size and approved services, but as of June 11 those guardrails are not yet in place. That gap matters. "Soon" is not a risk management framework.

Coinbase's response, per TechCrunch, is that users can restrict the agent to a sandbox account rather than their primary account. That's a meaningful option. But it depends entirely on users understanding the risk architecture well enough to make that choice deliberately, and most people probably won't.

Worth watching

Coinbase's announcement is timed into a crypto market that CNBC describes as "a relatively subdued, post-cycle slump." Bitcoin was trading around $63,461 as of data visible in the Decrypt source feed, well below recent highs. Launching aggressive agentic trading tools into a soft market is either a smart counter-cyclical move to deepen user engagement, or a way to generate fee volume by encouraging automated activity in an environment where discretionary human traders are sitting on their hands.

The x402 protocol has been live since May 2025, according to CNBC, and has been building out a list of compatible services. The key open question: whether regulators at the SEC or CFTC view autonomous agent-executed derivatives trading as something requiring a different disclosure or suitability framework than a human clicking "buy." Neither agency has announced guidance on that question as of June 11.