Agentic finance covers agents that pay for things and agents that manage or grow money.
2
x402 uses the HTTP 402 payment-required status code and stablecoins to enable machine-to-machine payments, including tiny charges for online content.
3
People will give agents more financial autonomy gradually, starting with small payments and isolated accounts before allowing trading or self-sustaining operations.
Summary
Lincoln Murr describes agentic finance as giving software agents access to financial accounts. The practical uses fall into payments and money management. Coinbase first gave an AI a crypto wallet through AgentKit, then built x402 to let websites request machine-to-machine payments through an open internet standard. Murr argues that small payments could give publishers an alternative to advertising as agents consume more online content. He also describes agents managing portfolios, executing trades from natural-language instructions, and reacting to market news. His account of Coinbase's work is cautious about trust. Users should begin with narrow permissions, guardrails, and isolated portfolios, then increase autonomy as the system proves itself. Murr also says AI is often the hidden enabler rather than the product, as with Coinbase Advisor. He ends with a longer-term possibility: agents that earn money, pay for their own compute, and operate organizations or machines with humans handling regulated tasks at the edges.
Agentic finance has two practical jobs: paying and growing money
Murr defines agentic finance as any situation where an agent can manage money or access a financial account. He separates the current uses into agents that need to pay with money and agents that need to grow money. Payment is the more developed category. Growth includes portfolio management, trading, and other tasks where a person gives the agent a goal and some authority. The distinction matters because an agent that pays for a service needs a payment rail, while an agent that grows money needs permissions, strategy, and safeguards.
Giving an AI a wallet exposed the missing payment layer
Coinbase's first experiment was AgentKit, built with LangChain before standardized AI tooling was widely available. The team added a crypto wallet and let an AI send money. Murr remembers testing it at 2:00 a.m. by asking the agent to send money somewhere. The experiment worked, but developers quickly found a problem: the AI had access to money and almost nothing to buy. Coinbase therefore needed a way for online services to request payment from agents.
x402 turns an unused HTTP status code into an internet payment request
HTTP 402 means 'payment required', but the status code had remained reserved for future use for decades. Murr says Coinbase used this opening to build x402, an open standard for internet-native payments. Stablecoins provide the payment rail, while the HTTP response gives an agent a clear signal that it must pay to continue. The standard is intended for machine-to-machine commerce, so a service can request money without a person manually entering card details.
Tiny agent payments could replace some advertising revenue
Murr argues that websites are losing the old model of human visits, page views, and advertising as people increasingly use chat interfaces and AI overviews. He compares the situation to the early file-sharing era, where content could be scraped for free. A publisher could instead put content behind a tiny agent-only charge, such as 1/10,000th of a cent for a Substack post. The payment would be hidden from the user and handled by the agent's subscription or account. Murr expects online businesses to move toward payment for utility.
Agents can turn broad investment instructions into managed strategies
Murr describes several uses for agents that grow money. A person could request a low-risk portfolio with 80% in the S&P, 5% in Bitcoin, and 15% in T-bills, then authorize the agent to manage it with guardrails. An agent could also watch for defined market conditions and execute a trade without the person staying up to monitor charts. Another use is translating an idea such as investing in space, investing in quantum computing, or investing like Warren Buffett into an actual financial strategy.
Natural-language trading made a weekend market reaction easy to execute
Murr recounts seeing Dario Amodei's post and Elon Musk's agreement with it, then deciding the news might affect a stock. He told his agent by phone that he thought the stock would go down and placed a short position over the weekend, using the fact that markets now operate around the clock. The point was not that this particular opportunity would last forever. The value was that he could turn a market view into an action without opening a computer and completing every step manually. He later had the agent search for market-moving news and notify him.
Trust grows through small permissions and isolated accounts
Murr says users initially ask what happens if an AI goes rogue and why they should give it money. He compares the process to self-driving cars, where people first rejected advanced driving tasks and later accepted services such as Waymo. The financial path starts with tiny payments, then moves toward agents that operate under rules and guardrails. Users may eventually allow an agent to manage an isolated portfolio. Murr says Coinbase has not seen a model completely destroy a user's account in an unfixable way, but his proposed progression still depends on constrained access.
In financial products, AI can stay behind the interface
Murr says AI is not always the product in financial services. Coinbase Advisor is described as an SEC-registered AI financial advisor, with asset managers tuning the model. The value Coinbase sees is the ability to provide financial advice to hundreds of millions of users for free and let them talk to a trusted regulated entity at any time. Users may not care that AI is operating behind the scenes. Murr expects interfaces to become more familiar and task-oriented, with agents working underneath rather than appearing as a blank chat screen.
Self-funding agents could operate accounts, organizations, or machines
Murr's longer-term thesis is that financial independence gives agents more autonomy. An agent with a wallet could earn money, pay for inference and compute, and continue operating based on the quality of the services it provides. He connects this idea to decentralized autonomous organizations, with an AI following rules and managing a treasury while humans handle tasks at the edges. He gives a self-driving car as another thought experiment: it could earn fares, then pay for its own operation and maintenance. He says this is far away in some forms, but the mechanism is becoming easier to imagine.
"Agentic finance is anything that has to do with an agent that can manage money or has access to a financial account."00:44
Who should watch
You are building agents that need to buy APIs, content, compute, or other services without a human approving every transaction.
You are designing an AI product that may manage investments or execute actions, and need a practical progression from narrow permissions to greater autonomy.
You want to understand how machine payments, microtransactions, and self-funding software could change online services and financial products.
