The story being told about x402 right now is that it ends the subscription era and finally makes fractional-cent payments work for everyone. I want to push back on that, not because x402 is unimportant, but because the framing repeats a thirty-year-old mistake.
Micropayments have been predicted, launched, and buried repeatedly since the mid-1990s. They did not fail on technology. They failed on human psychology.
What is genuinely new about x402 is not the rail underneath it. It is that the party making the payment decision is no longer a person.
The Thirty-Year Graveyard
Digital Equipment Corporation built Millicent in the mid-1990s, a script-based system explicitly designed to support payments as small as one tenth of a cent. CyberCash and its CyberCoin product handled small online payments before the company’s assets were sold to VeriSign in 2001.
Beenz and Flooz, the two best-funded consumer digital currencies of the dot-com era, both collapsed in August 2001 within days of each other. Flooz went bankrupt after unknowingly selling around $300,000 of currency to a Russian and Filipino organised crime ring using stolen cards. These were not engineering failures.
Later attempts kept the pattern going with better technology. Brave’s Basic Attention Token, launched in 2017, tried to reward attention and let users tip creators in BAT, but it never displaced advertising as the browser’s economic model. Lightning Labs built L402 (originally LSAT) on top of Bitcoin’s Lightning Network, reviving the same dormant HTTP 402 status code to meter API access per request.
The technology worked. Adoption for human-facing micropayments did not follow.
The reason was diagnosed before most of these ventures even launched.
Nick Szabo’s 1999 essay “Micropayments and Mental Transaction Costs” argued that the cost of deciding whether something is worth a tiny sum would soon dominate the technical cost of the payment itself. Clay Shirky put it more bluntly in “The Case Against Micropayments” (2000) and again in “Fame vs Fortune: Micropayments and Free Content” (2003): users hate them, the weakness is systemic, and so they will keep failing.
The point both men made is that there is no such thing as a no-brainer transaction. If a charge is large enough to be worth collecting, it is large enough that a human wants to think about it, and that split-second of thought, repeated across a day, costs more than the money involved. Flat-rate subscriptions and advertising win every time a human is in the loop, precisely because they remove the decision.
Why x402 Is Actually Different
Because it’s reframing.
Every prior scheme tried to lower the technical cost of a small payment. Szabo’s insight was that the technical cost was never the binding constraint. The binding constraint was cognitive, and it lived in the buyer’s head. x402 does not solve the cognitive problem. It removes the buyer’s head from the transaction entirely.
An AI agent does not have mental transaction costs. For an agent, deciding whether to spend a tenth of a cent on an API call is arithmetic under a budget constraint, not friction.
It does not feel like hassle, anxiety, or decision fatigue. It does not need reassurance that it is not being overcharged.
The one barrier that killed Millicent, Beenz, Flooz, BAT, and human-facing L402 is dissolved, and it is dissolved as a side effect of agentic commerce, not by anything clever in the protocol. This is why the stablecoin rails, while necessary, were never the missing piece. Cheap, fast, programmable settlement is a precondition. It is not the reason this moment is different.
The traction is where honesty is required: In the 30 days before the launch, x402 recorded about 75.4 million transactions moving only around $24.2 million, between roughly 94,000 buyers and 22,000 sellers, for an average payment of about 32 cents.
That average is the story. It is a couple of orders of magnitude above the sub-cent micro-economy the headlines describe.
A meaningful part of the early volume was not machine-to-machine metering at all. In October 2025, a meme coin called PING, minted by making a one-dollar x402 payment, turned the protocol into a speculative game: Dune data shows weekly transactions jumping 492% to a record in that single week, and PING’s market capitalisation briefly topped $57 million on October 25 before falling more than 40% within a day.
Chainalysis found that wallet retention cratered from around 87% to 5% once the speculative catalyst faded. The rails were being exercised by speculation, not by agents buying data.
Recommendations
From a seat running a regulated payments infrastructure, I would treat x402 as a real primitive for one specific job and resist the wider narrative. My advice is staged.
First, watch the right metric. Ignore cumulative transaction counts, which speculation inflates, and track dollar-volume concentration and repeat, non-speculative wallet activity instead.
The Chainalysis shift towards the dollar-and-above tier is the kind of signal that matters. If genuine machine-to-machine volume in that tier keeps rising while meme coin spikes fade, the thesis is holding. If not, it is noise.
Second, treat removing the human as removing judgment, not just friction. There are no native chargebacks, refunds, or dispute resolution, and on-chain settlement is irreversible.
That is a feature for a machine paying for a data call and a serious problem the moment a mistaken or compromised agent spends against a budget. The controls that used to live in a human’s hesitation have to be built into the system instead: agent spending caps, per-counterparty and per-period budgets, and allow-lists become the new control plane.
Third, do not assume compliance obligations shrink because the buyer is software. Anti-money-laundering and transaction-monitoring duties do not disappear when a payer is an agent. If anything, attributing a payment to a responsible legal entity gets harder, and irreversibility raises the cost of getting it wrong. Anyone deploying this at scale should assume the monitoring and identity questions arrive with it, not after it.
Caveats and the Measured Prediction
Subscriptions will not die wholesale. Humans remain in most consumer purchasing loops, and for them, Szabo and Shirky still hold. Expect hybrid models rather than pure per-call settlement: prepaid sessions, deferred vouchers, and batch settlement, because settling every sub-cent call on the chain does not scale economically once gas is a meaningful fraction of the payment.
The plausible near-term market is machine-to-machine API metering, agents paying for inference, data, and compute, not people paying per article.
x402 is a well-designed piece of infrastructure that finally gives the web a native payment primitive, and it arrives at the moment a buyer exists who does not mind making the decision.
That is the genuinely interesting part. It is not the death of subscriptions, and it is not, yet, a fractional-cent economy. Treated as machine plumbing rather than a consumer revolution, it is worth building for.
Disclaimer: This is a guest opinion piece. The views, thoughts, and opinions expressed herein belong solely to the author and do not necessarily reflect the official views or position of BeInCrypto or its editorial staff. This article is provided for educational and informational purposes only and does not constitute financial, investment, or legal advice.
The post The Web’s Missing Payment Primitive Didn’t Need Better Tech, It Needed to Ditch Humans appeared first on BeInCrypto.
