The Commoner

Monday, September 28, 2026

Sup, freaks.

Apollo's Torsten Slok is worried about AI agents taking cheap deposits away from banks. I think this goes well beyond finding a better rate. We already have services that let people save in bitcoin and pay their dollar bills. Agents could take a lot of the remaining work out of doing both.

WHAT TO EXPECT IN THIS NEWSLETTER

• Marty's Bent: saving in bitcoin while paying dollar bills, and how agents could take the work out of moving between the two.
• A full week of bitcoin ETF inflows and my latest conversation with Jordi Visser.
• PubKey's pro-data-center party, Bitget's withdrawal restart, and a closer look at the Shielded Bitcoin paper.
• Keep your phone number out of new Signal conversations, plus today's bitcoin and on-chain data.

Marty's Bent

The Savings Currency of the Agentic Economy

Torsten Slok, Apollo's chief economist, put out a note yesterday asking whether an agentic bank run is coming. People leave money sitting in checking accounts that pay next to nothing even though they could earn more somewhere else. Give those people an agent that can find a better account and move the money for them, and the banks could lose a significant amount of the cheap funding they've grown accustomed to.

An agent doesn't have to put the task off until next weekend. If it can do this reliably with its owner's permission, the bank has to compete a lot harder to keep that deposit. Moving deposits between banks doesn't remove them from the system, but it can make life very uncomfortable for the bank that was counting on its customers staying put.

Stablecoins look like they've won the first round of payments. Stripe already has infrastructure for agents to pay for services with stablecoins, alongside cards. Businesses price things in dollars, people have dollar bills to pay, and agents will use the payment methods their counterparties accept. I don't think bitcoin needs to win every one of those transactions to become the money people want to accumulate.

As agents begin doing more business with each other, I expect money to move much faster than it does in an economy where a human has to approve every little purchase. An agent buying compute or paying for data should be able to shop around while it works. The same should be true of the cash it keeps available. Banks offering less for comparable deposits will have to pay up or watch the money leave. Different risks will still command different returns, but getting paid simply because your customer couldn't be bothered to move should become a much worse business.

And once an agent has found the better dollar yield, it still owns dollars. Say you're earning five, six or seven percent somewhere. Those are more units of a currency that can be created at will. Whether you are actually getting ahead depends on what those dollars can buy when you spend them. I expect the debt problem to be met with a lot more money creation over time. If you're putting away money you worked hard to earn, that problem doesn't disappear because you've found a better rate.

Bitcoin has a predictable issuance schedule and a supply cap that doesn't adjust because a government has borrowed too much. Bitcoin's price will still swing around. For the surplus people want to put away for years, I expect it to become the obvious savings account.

We can already do a lot of this. Strike, River and Cash App let people receive direct deposits and automatically convert a chosen portion of their paychecks into bitcoin. Strike's Bill Pay handles the other direction. Select your bitcoin balance as the payment source and it sells bitcoin to cover your dollar bills when they're charged. Your credit card company doesn't need to accept bitcoin for you to save in it.

These services have made living on bitcoin significantly easier. Paycheck conversions already happen automatically. With Strike Bill Pay set to use bitcoin, you don't have to make a separate sale to cover a supported bill. But you still have to connect the accounts, choose the settings, make sure money is available when a bill arrives, and change things when your income or expenses change. Strike's own instructions warn that an automatic withdrawal to self-custody can leave too little available for a bill. Even with good automation, somebody has to make sure the different pieces work together.

Now give an agent reliable, permissioned access to those services. You tell it what needs to be paid, how much you want available for expenses, and that you want to save the rest in bitcoin. Within the limits you approve, it could coordinate incoming pay, conversions and bill payments, adjusting as your bills change instead of leaving you to manage a collection of fixed settings. That could reduce the dollars people leave sitting around simply because managing the transfers themselves is a pain in the ass. You could save in bitcoin without making account management a hobby.

When Jordi Visser joined me last week, he talked about investing in speed rather than friction. Long speed, short friction. This is a very practical application of that idea. We have people who want to hold bitcoin and bills that still need to be paid in dollars. Make moving between the two faster and easier, and more people can act on that preference. They don't have to wait for their employer, landlord and every merchant they use to adopt bitcoin first. I think removing that work is going to be massive for adoption.

I would not be shocked if we wind up with bitcoin sitting on both ends of a huge amount of agent commerce too. An agent converts what it needs into a dollar stablecoin, pays for a service, and the agent receiving the payment converts what it doesn't need for its own expenses back into bitcoin. Dollars become a hot currency. Useful while a transaction is in motion, held for as little time as the business can manage. The agent still has to account for conversion costs, taxes and settlement times. It can do the coordination without making those constraints disappear.

If you're earning a return on the dollars you keep available for expenses, have the agent sweep it into bitcoin. You don't have to keep accumulating the currency just because a provider pays you in it. Nor do you need to lend out your bitcoin to do this. The interest is earned elsewhere and used to buy more.

Bitcoiners have been describing this behavior for a long time. Spend the bad money and save the good money. Slok is worried that agents will get people out of bad checking accounts. I think they'll eventually help people get their long-term savings out of the dollar altogether.

SIGNAL

Another full week of ETF inflows

US spot bitcoin ETFs took in $2.39 billion net last week, according to our ETF flow tracker. Every session from September 21 through September 25 was positive. Including the two preceding positive sessions, the seven-session total comes to $2.98 billion.

The daily inflows got smaller as the week progressed, so this was sustained buying rather than an acceleration every day. Still, that is a substantial amount of net demand coming through brokerage accounts.

Jordi Visser on what agents will change

If you missed my conversation with Jordi Visser last Thursday, give it a listen. We got into what happens when agents become customers and tokenized assets become easier to use as collateral. Jordi's argument is that software can make capital more useful without people having to sell everything they own to access it.

That could happen while existing businesses have a very difficult time defending their profits. An expanding economy doesn't mean every company keeps its pricing power when customers can compare and switch providers more easily. Jordi picked up that theme again in yesterday's A Bear Market Inside a Bull Market.

PubKey throws a party for building things

PubKey DC hosted Data After Dark on Saturday, a pro-data-center party organized by Pirate Wires and Young Voices. The venue's weekend video gave people plenty to argue about. I like seeing people make an affirmative case for building American compute instead of treating another proposed facility as something we should automatically stop.

The case gets much stronger when the people living nearby can see what they get out of it. Lower property taxes or better services are worth explaining. So is who pays for the power infrastructure. Households shouldn't be handed the bill for a dedicated upgrade a developer needs. Build the capacity, pay for the costs you create, and give the town a reason to want you there.

Bitget says bitcoin withdrawals have restarted

Bitget says native bitcoin withdrawals resumed at 08:00 UTC today as it begins a phased reopening after its September 24 security incident. The exchange now puts the assets transferred to attacker-controlled addresses at approximately $387.5 million. It says the increase from its initial estimate reflects previously uncounted assets, not another round of unauthorized transfers.

Customers have had to wait for the exchange to restore access to their money. That is part of what you take on when you leave bitcoin with a custodian, even if you never intend to trade it. A balance on an exchange and bitcoin you can move with your own keys are very different things when withdrawals stop.

A closer look at Shielded Bitcoin

We mentioned the Shielded Bitcoin paper on Friday. Clara Shikhelman, Mikhail Komarov and Aleksei Moskvin propose Zcash-inspired private transfers without changing bitcoin's consensus rules. Encrypted notes hide amounts, while zero-knowledge proofs establish that a spender controls valid notes and isn't creating extra value. Public spend markers let the software catch double-spends without pointing directly to the earlier notes being spent.

Bitcoin would record and order the data. Separate software would check the proofs and reconstruct the private-transfer state. So this would run above bitcoin's existing consensus, rather than make ordinary onchain transactions private.

The paper leaves entry and exit outside its scope, pointing to a separate PIPEs v2 design. Its proof system also requires a trusted setup, and timing, fees and other metadata remain visible. There is real work left before this becomes something people can use. But hiding amounts and payment relationships while using bitcoin's ordered record is research worth following.

⚡ FREEDOM TECH CORNER

DATA SNAPSHOT

Retrieved September 28, 2026, 8:26 a.m. ET. Kraken last trade and mempool.space network readings are retrieval-time observations. Fear & Greed is dated September 28 UTC. Bitcoin Lab MVRV, realized price and SOPR are daily observations dated September 27 UTC.

Sources: Kraken for last-trade price; mempool.space for network estimates; Alternative.me for daily sentiment; Bitcoin Lab for MVRV, realized price and SOPR. Sats per dollar is calculated as 100,000,000 divided by the Kraken price. MVRV compares market capitalization with realized capitalization. Realized price is realized capitalization divided by circulating supply. SOPR compares spent-output value at spending with value at creation; above 1 indicates aggregate realized profit. Bitcoin Lab values are reported as supplied, not calculated from the live Kraken price.

If this issue gave you something to think about, forward it to a friend and ask them to subscribe to The Commoner.

See you tomorrow,

Marty

Marty Bent · TFTC · Nostr

YouTube

Podcast