Coinbase Puts AI Agents On The Trading Desk With x402 Payments

Coinbase launched a tool this week that lets AI agents trade crypto, move money and pay for online services through its x402 payments protocol within preset spending limits.

Key Points:

Coinbase For Agents Goes Live

The product is a standalone account rather than an in-app feature, and it connects to a user's existing Coinbase balance only after spending limits are set.

Assistants like ChatGPT and Claude plug in through a server. From there an agent can rebalance holdings across Bitcoin (BTC), Ether (ETH) and Solana (SOL), or buy dips through preset price moves without funding a separate wallet.

Users decide how much rope to give, from a single recommendation, to one automated trade a day for a week, to a standing thesis the agent runs on its own. Equities and prediction markets are slated to follow.

Also Read: Cardano Whales Roar Back To Life As ADA Tests Multi-Year Lows

x402 Powers Agent Payments

Payments run on x402, an open standard Coinbase introduced last year, letting agents buy paywalled research, data feeds, premium APIs and on-demand compute with no login or subscription. Stablecoins are the preferred rail, with USDC (USDC) settling most transactions. In one flow, an agent pays a few cents for data behind a paywall, reads it, then places a trade off the result, and Coinbase says the same rail can later cover shopping and checkout flows.

Why The Coinbase Bet Matters

Lincoln Murr, Coinbase's head of AI product, casts the launch as an edge over pure trading apps, since it pairs exchange access with a native payments rail few rivals can match. The aim is to own the account layer that AI assistants route through.

It is a wager that the next financial interface is a chatbot, not an app.

An agent can run inside a walled-off sandbox with no view of a user's other holdings, or draw on the main account only up to the spending caps that a user sets. Coinbase likens the setup to a budget card rather than a bank account, a hedge against a model that misreads a signal, follows a bad prompt, or is steered by a malicious instruction.

The release caps a multiyear push that began with AgentKit in 2024, followed by an in-app assistant that offered trading tips and advice last December.

It arrives just days after Robinhood opened trading to its own agents, part of a broader race into agentic finance that now spans exchanges, brokerages and payment networks. Regulators are watching, too, after the Financial Stability Board last week urged firm safeguards around the rise of agentic AI in finance.

Read Next: OpenAI Targets Anthropic With Price Cuts Ahead Of A Pivotal IPO