Could an AI agent quietly moving your savings around actually shrink the whole banking system's deposits — or just shuffle them between banks?

This explores which kinds of money movement actually shrink the total pool of deposits across all banks, as opposed to just shifting money from one bank to another. The collection mostly answers by ruling things out: it shows clearly what does *not* drain the system, but it never lists the channels that do. The core argument is accounting. When an AI agent moves your savings from Bank A to a higher-paying Bank B, the deposit just changes address. Bank A loses it, Bank B gains it, and the system total stays the same Would AI cash-sweeping agents trigger a systemic bank run?. So the fear that rate-chasing agents could empty the banking system mixes up money moving *between* banks with money leaving the banking system altogether.

The risk that remains is narrower but real. Cullen Roche concedes that agents could turn a withdrawal that used to take days into one that takes hours. That speed could finish off a single bank with a weak balance sheet in the style of Silicon Valley Bank, even though the system as a whole stays intact Could AI agents speed up runs on individual weak banks?. McKinsey's version of the threat is about profits, not survival. Banks earn money because customers don't bother moving to better rates. If 10 to 20 percent of consumers let agents sweep their cash, deposit margins could shrink by 30 to 50 basis points Will AI agents break the deposit and credit card inertia?. So the realistic effect of agents is that deposits get more expensive for banks to keep, not that they disappear.

That business model has a close parallel in subscription billing. Todasco argues that agents will undermine negative option billing (where you're charged until you cancel) by removing the forgetfulness that let inattentive customers subsidize the attentive ones Will AI agents destroy the negative option billing business model?. Idle low-interest deposits work the same way. In both cases, the agent's job is to wipe out the profit businesses make from customer passivity.

The gap: the corpus doesn't spell out the channels that really do take deposits out of the whole system. In standard monetary mechanics, those are the moves where money leaves for somewhere other than another bank. Examples are buying Treasury bills or money-market funds that park cash at the Federal Reserve, paying taxes into the Treasury's account at the Fed, paying down loans, withdrawing physical cash, and the Fed shrinking its balance sheet. The question worth asking is therefore not whether agents move money, but whether agents would ever send sweeps to one of those destinations instead of to another bank. The corpus doesn't test that, and it's the scenario that would make the systemic-run worry come back.