machines can't open a bank account

an investor told me he doesn't care which payment rail wins the machine economy. maybe ai systems will just barter compute. here's why i still think the answer is bitcoin.

share x linkedin facebook

a small boxy robot stands on a warm field facing a tall grey brick wall, while a glowing gold coin circled by dotted orbit rings hangs in the pale sky above

in the same monday taping, i asked the investor about the machine economy. robots and ai agents paying each other for compute and energy. it's the thesis i've been building around for years, so i was curious where he'd land.

at first he heard my question as something else. he talked about ai making businesses cheaper to run, needing less capital. that's true, but it wasn't what i asked. so i pushed. i don't mean cheaper companies. i mean machines that need a native way to pay each other, with no human in the loop.

once we were on the same page, he took a contrarian position. he said he doesn't care which rail wins. bitcoin, card networks, stablecoins, whatever. he thinks one way of exchanging value will emerge, and it might not even be money. ai systems could barter. i give you compute, you give me analysis.

i liked that answer because it forced me to sharpen mine.

barter is a real possibility between two machines that each have something the other wants at the same moment. that's the old problem with barter between humans too. it needs a coincidence of wants. the moment you have a million machines with different needs at different times, you need something that holds value in between. that's what money is for.

so the next step is to ask which money a machine would choose to hold.

a card network needs an account, a company, a person who signed an agreement and can be sued. a robot can't open a bank account. an agent spun up for six minutes to finish a task can't pass a know your customer check. most of the rails we have assume a human at each end.

a bitcoin wallet assumes nothing. a machine can generate a key and start paying other machines, anywhere, at any hour, with nobody's permission. on lightning it can do that in tiny amounts, fractions of a cent for a second of compute. and nobody running the network can print more of it, which matters when the machines start holding balances for longer than a moment.

stablecoins have a place in that picture too, and i've said so before. they're useful rails for agents that need to price things in dollars. but a stablecoin has an issuer who can freeze it. for machines that have to trust nobody, a neutral base layer matters.

he and i agree on more than it sounds. he's right that transaction costs keep falling and that the winner will be whatever is cheapest and simplest for machines to use. i think that ends up being the money that no company controls.

this matters for founders now, not in ten years. if you're building anything with agents, you'll hit the payment question sooner than you think. the agent that books or buys something needs a wallet, so design for it from day one.

long term, i think the machine economy settles on the money machines can hold without asking. we'll see which of us was right.

i wrote the longer version of this argument in the machine economy book.

so if your product had an ai agent as a customer tomorrow... how would it pay you?

share x linkedin facebook
the newsletter

the machine economy brief

one email when it matters: bitcoin, ai, robotics, and what founders should do about it. unsubscribe anytime.

no spam, no list-selling. your email goes to sunny, nowhere else.

raising right now?

i run fundraising as a service. lead-investor manufacturing, warm intros, the whole motion.

see how raasrocket works →