Your Checking Account Runs on Friction

Ethan Mollick keeps coming back to a point I think about a lot. Even if AI development stopped today, the models we already have would take years to work through the economy. He's put it anywhere from five years to a couple of decades. He calls the gap between what the models can do and what we're using them for the overhang.

I've mostly thought about the overhang in terms of work, meaning how a team writes, researches, and builds software. In late September, a post from Jason Yanowitz pointed me somewhere I hadn't looked.

Chief Economist at Apollo: agents could cause a bank run by sweeping household cash into accounts paying 3-5% instead of the 0.1% national average, causing banks to lose a large share of their cheap deposits.

The economist is Torsten Slok, and the backdrop is Meta's Muse, a personal agent that connects to your financial accounts, watches balances, and acts on your behalf. Bank stocks dipped when it launched. A Bank of America analyst called "deposit sorting," the frictionless movement of excess cash to higher yields, a real threat to bank margins, while noting it's still conceptual until deposit costs start rising.

Friction is part of the business model

A lot of our economy is built on friction. Nobody picks a checking account paying 0.1% because it's a good deal. We leave cash there because moving it around is a pain and managing it moment to moment is worse. We don't want to miss a payment because the money was sitting in the wrong account. The low rate is what we pay to not think about it, and banks have collected that payment for decades.

Companies don't work this way. A corporate treasurer sweeps idle cash into money market funds and pulls it back when bills come due. That's worth doing when the balance is large and watching it is someone's job. For a household, the effort has never been worth the yield.

An agent changes the effort side of that math.

Moving money inside one bank

After reading the post, I searched for the bank that's most API-forward, the one an agent could work with most easily. The answer was Mercury, which I already use for my business account.

An API-forward bank covers the simplest version of this. An agent could keep checking near zero, sweep the rest into savings, and move it back a day before a payment goes out. The bank already allows every one of those transfers. What's been missing is someone willing to watch the balances and the bill dates closely enough to do it without ever missing a payment.

Getting my personal money there is its own problem. Changing accounts is miserable because of everything wired to the old one. Every auto-withdrawal, every direct deposit, every service that bills you has to be found and updated, and missing one means a failed payment. That work is what keeps money parked. If an agent could chase all of it down, money that's effectively static becomes liquid.

Moving money between banks

A savings account is the floor, though. Chasing better returns means going to other banks, money market funds, or brokerages. Crypto platforms are in the mix too. Robinhood Earn advertises an estimated 7% on a dollar-pegged stablecoin, with risks a savings account doesn't carry. The more easily money moves, the more it can go wherever the return is best for the risk you're willing to take. All of that means moving money between institutions, and that's where the transfer itself gets slow.

Most bank-to-bank transfers still run on ACH, a system that settles in batches. A standard transfer takes a day or more, stops for weekends and holidays, and misses the day's window if you start it after the cutoff. Instant payment rails exist now, but plenty of banks haven't turned them on. When cash takes days to arrive, you keep a cushion where the bills get paid, and that cushion earns 0.1%.

Some of that is old technology. Some of it suits the banks fine. Money that's hard to move tends to stay put, and deposits that stay put are cheap. So you get transfer limits, holds on incoming funds, fees on outgoing wires, and account links that break and need re-verifying. Each is a small annoyance with a plausible reason behind it. Together they build a wall around your balance, and the bank has little reason to lower it. An agent doesn't make ACH faster, but it can plan around the delay, starting the transfer back days before the bill instead of hours.

Households managing cash like banks

With APIs to move cash and agents to pull the levers, individuals could start managing cash the way companies and banks do. Sweep the float into whatever pays best for the risk, pull it back before the mortgage hits.

Maybe. The amounts sitting in most household accounts are small enough that the extra yield may not be worth the risk or the trouble. A few hundred dollars a year doesn't cover much if an agent mistimes a transfer before rent is due, and a 7% crypto yield carries risks an FDIC-insured savings account doesn't.

Still, the thing keeping this from happening has been effort, and effort is what agents remove. First it becomes possible. Then some people do it. Eventually it's a default setting in whatever app manages your money, and banks' cheap deposits stop being cheap.

Where the overhang shows up

This is what I think Mollick's decade looks like up close. Muse runs on models that exist today, so none of this waits on a smarter model. What takes time is finding every place where friction was quietly holding something still, and watching it start to move.