The Ghost Rails

China built entire cities nobody moved into. Highways to nowhere, towers with no lights on. They weren't failures of construction. They were failures of timing: infrastructure waiting on users.

Jordi Visser thinks crypto is a ghost city.

Fifteen years, billions of dollars, and what got built? Lending markets with no loan officers. Stablecoins that settle in seconds with no bank. Tokenized assets you can slice into fractions no wire transfer can match. Enormous, elegant plumbing. And the users never really showed up. Retail arrived in waves and left in crashes. Institutions dipped a toe and called it allocation. The rails kept getting built anyway, on faith.

Visser’s argument, laid out in a Bitcoin Magazine interview last week: the faith was misplaced only about who the users would be. The rails were never built for us.

Look at what crypto actually built, and ask who it's for.

Stablecoins solve a problem humans barely have. You and I can wait two days for a wire. A software agent running a thousand transactions a minute cannot. Lending protocols run on code instead of credit officers, which matters enormously if the borrower has no identity, no credit score, and no pulse. Tokenization lets value move in slices and at speeds that only make sense at machine scale.

Every one of these primitives gets more valuable the less human is in the loop. An agent doesn't need an app interface or a support line. It needs deterministic execution, cryptographic proof, and settlement that never sleeps. Crypto is the only financial system that checks all three boxes.

And the machines are already knocking. Coinbase's x402 protocol revived a piece of the web that sat dormant for thirty years: HTTP status code 402, Payment Required. The flow is beautifully simple. An agent requests a resource, the server answers 402 with a price, the agent signs a stablecoin payment and retries. No account. No API key. No human clicking confirm at 2 a.m. Your wallet is your identity.

Visser's historical analogy is the one that sticks. Fourteen years passed between Netscape going public and the App Store putting the internet in everyone's pocket. Infrastructure first, users later, and nobody in 1996 could picture the user that finally arrived. This time the user isn't holding a phone. It runs around the clock.

Now the cold water.

The traffic on these rails is still a trickle. TRM Labs went through x402's settlement data and screened roughly $52 million of volume down to about $25 million of plausibly genuine commerce. Of that, the genuinely agentic activity sits somewhere between $154,000 and $1.9 million. Total. Since launch. That is not an economy. That is a demo day.

Worse, agents don't need crypto. Google, Stripe, and OpenAI are all building agent payment rails on the fiat system, and they have something crypto doesn't: distribution. An agent paying through Stripe's infrastructure never touches a blockchain. The ghost rails have well-funded competitors, and the competitors own the on-ramps.

And there's a deeper problem with the thesis. The users are coming is what every ghost city planner believed, too. Maybe the rails weren't built for machines. Maybe they were built for humans who never came, and the agent story is just the newest explanation for fifteen years of failed adoption. Infrastructure in search of a user is indistinguishable from infrastructure nobody wants, right up until it isn't.

Here's where it gets personal.

I work in ad tech, and I watch machines move money every single day. Algorithms already buy and sell ads at superhuman speed. They just do it on Visa's rails, through corporate accounts, with humans reconciling the invoices. The jump from algorithm with a company credit card to agent with a stablecoin wallet is not a leap. It's a paperwork change.

That's what makes Visser's thesis land for me. The question was never whether agents will transact. They already do. The question is which rails they choose when the training wheels come off, and whether the open, programmable, always-on system beats the closed one with better distribution. I've spent five years watching the closed system win in my industry. It doesn't always win.

So what do you actually do?

Stop watching crypto Twitter for adoption signals. The users Visser is describing don't tweet. Watch stablecoin supply, watch x402 and agent-protocol volumes, watch who builds the facilitators and wallets. When the machines arrive, they won't announce themselves. They'll just show up in the data.

Position-wise, the bet is simple: machine demand is still demand. You don't have to pick which agent wins or which protocol becomes the standard. You bet that autonomous software needs a monetary layer that's open, programmable, and always on, and you own the reserve asset of that layer while it's still priced for human adoption.

The ghost city doesn't stay empty forever. Somebody always moves in.

If this made you think, forward it to the biggest crypto skeptic you know and the biggest AI optimist you know. Then grab popcorn.