AI agents will no longer let money sit idle
Today, many customers still keep money in checking accounts that pay almost no interest. This inertia is valuable to banks. But AI agents are beginning to become capable of managing wallets and payments without human intervention.
Apollo provides a telling example. The average yield cited for US checking accounts is around 0.1%, while Revolut, SoFi, Varo, LendingClub and Wealthfront offer between 3.3% and 5% on certain products.
On $10,000, the difference amounts to approximately $10 in annual interest versus $330 to $500.
A human may put off switching banks for months to avoid the paperwork. An AI agent has no such problem. It can monitor rates continuously, automatically move available cash to the highest-yielding option, then transfer just enough money back to the main account before the rent or a bill is due.
In his note titled “Is an Agentic Bank Run Coming?”, Sløk estimates that if this behavior becomes widespread, banks could lose a significant share of the low-cost deposits they use to fund their loans.
No need for panic outside bank branches. A few lines of code could be enough.
The bank run could become algorithmic
The term “bank run” may sound excessive. The scenario described by Apollo does not necessarily resemble a classic bank run, in which customers withdraw their savings because they believe their bank is in trouble.
It is more like an automated migration of cash.
Banks could respond by increasing the interest paid on deposits. That would address part of the problem for customers, but increase banks’ funding costs and potentially reduce their margins.
BrefCrypto had already observed that Visa is preparing payments made directly by AI agents. The new element is that an agent will probably do more than simply make payments. It could move funds between multiple accounts, assets and institutions.
Coinbase has specifically strengthened this model with Coinbase for Agents. Since September 22, an agent has been able to access stocks and ETFs, monitor certain market conditions, purchase data and then execute a transaction according to limits set by its user.
The x402 protocol takes automation even further. Coinbase says it has already processed more than 230 million transactions for $54 million in volume in just over a year.
Machines are therefore genuinely starting to handle money.
Crypto already has the rails for agents
This is where the banking story directly meets crypto.
An AI agent operates 24 hours a day. It may want to move a few cents or several thousand dollars without filling out a form, waiting for a bank to open or entering a card number for every transaction. Stablecoins are particularly well suited to this model.
With x402, for example, an agent can automatically purchase financial data or an API call and pay directly in USDC. Coinbase also allows businesses to keep their balance in USDC, with a yield currently advertised at 3.35% for certain eligible users.
This competition could therefore come from two directions at once: fintechs that offer better returns on deposits and crypto infrastructure that makes money programmable.
Banks have understood this. Standard Bank, Goldman Sachs, Citi and other major institutions are already preparing their own stablecoin rails. What recently still looked like a blockchain experiment is beginning to affect their core business directly: holding customers’ money and moving it through the financial system.
Several obstacles obviously remain. Regulation, deposit insurance, agent security, execution errors and permissions will probably limit their autonomy. Giving an AI the right to move all of one’s savings is not a trivial matter.
Sløk’s reasoning is nevertheless difficult to dismiss. One of the major advantages banks still enjoy comes from human inertia. AI agents could eliminate precisely that inertia.
In this scenario, stablecoins and crypto would no longer simply be competitors to bank money. They could become the rails used by algorithms to constantly seek out the best place to put that money.
