I am writing a series of columns looking at the effect the sudden flood of agentic AI capabilities will have on the rest of us.
I wrote about the effect of AI agents on the advertising-supported online economy here, but there is another entirely different battle under way, in banking. It will affect us all, at least those of us with some money sitting in a deposit account somewhere.
In the early days of crypto decentralised finance there was a slew of “yield farming” products that sniffed the crypto space to find the project paying the most yield, immediately moving your money there. It was risky, required technical expertise and depended for its profits on the crypto markets being in a good mood. The rest of the world did not know or care about this tiny corner of crypto finance.
But yield farming is about to arrive in traditional finance, thanks to this new agentic AI beast (or swarm of beasts, to be more accurate). And it has burst into the financial zeitgeist with a note published by Torsten Sløk, chief economist at the US-based Apollo Global Management (one of the world’s largest alternative asset managers and a major player in the retirement services industry, with over $1 trillion in assets under management).
His note starts with a question: “Is an Agentic Bank Run Coming?”
His answer is alarming. Meta’s Muse and similar AI assistants, he argues, “could soon sweep household cash automatically into accounts paying 3.3% to 5.0%, instead of the 0.1% national average on checking accounts”.
“If every household used AI agents to optimize the return on their cash balances,” he writes, “banks could lose a large share of the cheap deposits they rely on to make loans, which would be a problem for the entire financial system”.
This has nothing to do with crypto anymore.
Consider this scenario – you simply say to your laptop (voice will do fine) – “please go and look for the most reputable financial institution which is paying the highest yield on a current account. If it is higher than that of my current bank, please close my account, open an account at the new place, and move my money there”.
While some may consider this a simplification of the complex and often tortuous process of closing and opening accounts (IDs and addresses and authorisations and the like), it is exactly what AI agents are good at.
It will just keep hammering away on your behalf until the new account is open, following whatever guardrails you set up (like “please let me choose the password, do not do it on my behalf”, or “prompt me when you need a copy of my ID”).
American banks hold roughly $19.6 trillion in deposits. About $4 trillion of that earns literally nothing – not a token rate, nothing – and a great deal more earns a rate that rounds down to nothing. Meanwhile, money market funds, Treasury bills and a dozen perfectly respectable fintechs will pay you something north of 3.5%. The gap between what your bank pays you and what your money is worth is roughly 350 to 400 basis points, and it has been sitting there, in plain sight, for years (at least in the US).
That gap is not an accident. It is the product.
The industry’s own phrase for it, from a Fintech Blueprint essay by Luke Spill, is the “inertia premium” – the margin earned from customers failing to compare rates or switch accounts.
It is the politest available description of a business model that depends on your being lazy and neglectful. Bank of America held $957 billion of consumer deposits at an average rate of 48 basis points (0.48%). Those depositors seem utterly uninterested in looking for a better deal, which would be trivial to find. The bank, meanwhile, makes ridiculously easy money on the gap between the tiny yield it pays us and the much higher yield it can extract from the same money.
The sums potentially exposed are enormous. McKinsey estimates that the world’s banks hold around $70 trillion in deposits, of which approximately $23 trillion sits in checking accounts earning little interest. Its modelling suggests that if agents caused only 5–10% of checking balances to migrate towards higher-yielding alternatives, total banking profits attributable to deposits could fall by 20% or more.
It would be a wholesale destruction of the machine that exploits depositor laziness.
The first destination for the money needn’t be anything remotely exotic. It could simply be another bank.
High-yield savings accounts, T-bills and fintech savings services already provide alternatives to leaving excess cash in a current account. Britain’s Riff, for example, describes itself as a “Savings Agent”. It is not live yet, but the pitch is this: using open banking, it will monitor customers’ balances and move savings between higher-paying accounts, keeping the money in government-insured institutions and, where necessary, spreading it among banks to remain within insurance limits.
The scenario that unfolds to this point suggests that banks should be panicking, wondering about the coming swarm of agents to sweep customer deposits away to more comfortable homes. What will they do? Raise their rates to something more competitive? It would be neat to end the story here, a fable of technology forcing a reckoning.
But alas. It turns out that consumers do not trust agents designed by tech companies or anyone else to move their money around, even with the promise of better returns. And so they will likely do very little with these tempting offers of agent-mediated yield farming. How do we know this? Because there is research to show it.
Two economists, Cirelli and Olafsson, studied what households do when offered a higher yield on equally safe money with instantaneous transfers – which is to say, the precise conditions an agent would create. Households moved 0.1 percentage points of their balances for every full percentage point of spread. Not ten per cent of it. A tenth of one per cent.
The agentic condition, in other words, has already been tested (at least in Iceland, where the study was done), and inertia won comfortably.
There is more. A service called MaxMyInterest has offered automated yield sweeping for about a decade, well before agentic AI, and almost nobody uses it. A TD survey in February found that 55% of Americans now use AI to help with their finances, up from 10% a year earlier, but only 18% would let it act on its own. Worse – it is the small accounts that seem to chase yield, while balances above $25,000 show no significant response at all.
Publicly, the big banks are relaxed to the point of boredom.
At a Morgan Stanley conference in June, PNC’s Bill Demchak said of the agentic threat, “I don’t understand where all that’s coming from.” On the evidence available, he was entirely correct.
Still, there are some tentative defences being tested in back rooms. Legal challenges, technical blocks, high API costs to deter machine switching of accounts, and other embryonic measures. And this – banks can also fight machines with machines.
If your agent decides to move your cash because another bank pays 4.4%, your bank’s AI can recognise that you are likely to leave and offer you 4.45%. McKinsey describes one bank using AI to identify excess customer liquidity, estimate the probability that it would depart, and selectively reprice deposits.
One can imagine where this ends. Your agent announces that Bank B is offering a better rate. Your bank’s agent offers another ten basis points. Your agent asks for fifteen. Somewhere in cyberspace two machines haggle over your savings while you are making coffee, unaware that you have just become a much better depositor.
The banks are going to rely on our inertia for as long as they can. The agents are going to be deployed anyway, chipping away at their defences. And then one day, without anyone really noticing it, banks will start raising their deposit yields to compete with the pesky but growing agent offerings. And then Sløk’s question will have become moot.
But it is not happening today.
[Image: Leo_Visions on Unsplash]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend