Software Is Buying Things Now. You Just Don't Control the Wallet Yet.

The average payment on x402 last month was 32 cents.

Nobody buys anything for 32 cents anymore. A coffee costs $5. A song costs $1.29. Even a single stock trade that would've cost you $7 in commission twenty years ago now costs nothing — because the brokerage makes it back selling your order flow for fractions of a cent per share.

But 32 cents? That's not a human price. That's a machine price.

I track where value settles, not where it gets spent, and when I see an average transaction value that low at scale, I know the buyers aren't people browsing and clicking "purchase." They're agents — software paying software for an API call, a model query, a slice of training data. No shopping cart. No credit card form. No human in the loop at all.

HTTP status code 402, "Payment Required," has sat in the web's technical specifications since 1997, reserved but never implemented. There were never any rails behind it. For three decades, the placeholder just sat there in the protocol documentation, a promise the internet never kept.

Now there are rails. And the 32-cent average tells you exactly what's running on them.

The Numbers Are Small. The Players Aren't.

Let's establish scale. Last month, x402 processed roughly 75 million transactions, totaling about $24 million, across 94,000 buyers and 22,000 sellers. In the world of payment networks, this is a rounding error. Visa processes 800 million transactions per day. Stripe moves billions.

But I learned a long time ago: don't measure disruption by current volume — measure it by who shows up early.

The x402 membership roster reads like a "who's who" of the payments and infrastructure world. Visa. Mastercard. American Express. Stripe. Google. Cloudflare. Circle. Forty members operating under the Linux Foundation umbrella, which is how you know this isn't vaporware or someone's side project.

Google has already wired x402 into its agent stack. Cloudflare ships it as part of its developer toolkit. When the card networks — companies whose entire business model is inserting themselves between buyers and sellers — voluntarily join a rail built to route around cards, they're not being altruistic. They've read the board. They know what's coming.

This is the railroad moment. Nobody gets fired the day the railroad arrives. The town just slowly empties out.

The Hard Part Was Never Throughput

I was on a call last week with a CFO trying to understand why his AI vendors' bills were so unpredictable. The models his team deployed were making API calls to external data sources he didn't even know existed. The usage was legitimate, but the spend was uncontrolled. "How do I set a limit?" he asked.

Good question. The answer right now is: you probably can't, not really.

See, when people talk about microtransactions or machine-to-machine payments, they obsess over throughput and latency. Can the system handle a million transactions per second? Can it settle in under 100 milliseconds? Those are solvable engineering problems. Stripe solved them. Visa solved them decades ago.

The actual product isn't the payment rail. It's the control surface: wallet policy, agent authorization, refunds, abuse limits.

When your agent is autonomously buying data by the API call, who sets the spending limit — you, your vendor, or the settlement layer they're using? If your model goes rogue or gets prompt-injected into buying garbage data at scale, how do you claw that back? What does a "refund" even mean when software made the purchase decision, not a human?

The companies building x402 aren't solving a speed problem. They're solving a governance problem. Whoever owns that control layer owns the settlement infrastructure of the machine economy. And right now, that's up for grabs.

32 Cents Is What Data Costs When Software Does the Buying

Let me connect this to something you've seen before.

In 2008, the New York Stock Exchange went fully electronic. For decades, humans in colored jackets had stood on a trading floor, shouting and hand-signaling to execute orders. Then algorithms took over. High-frequency trading firms started locating their servers as physically close to the exchange as possible — because at machine speed, three milliseconds of latency was the difference between profit and loss.

The shape of markets didn't just change. The entire concept of what constituted a "trade" changed. Humans think in minutes or hours. Machines think in microseconds. A human trader might make 50 trades in a day. An algorithm makes 50,000. The unit economics, the risk models, the regulatory frameworks — all of it had to be rebuilt from scratch because the buyer had changed.

That's what's happening now with payments, except the surface area is bigger. It's not just equities. It's every API, every model query, every data feed. Your software is already buying things. You just don't have the controls yet to govern how it spends.

When I look at that 32-cent average, I don't see a cute technical curiosity. I see the unit economics of a market where humans aren't the customers anymore.

The Questions You Should Be Sitting With

I'm not going to hand you a tidy action plan, because the honest answer is that most organizations aren't even asking the right questions yet. But here are the uncomfortable ones I'm sitting with:

If your AI agents are making purchasing decisions autonomously, where does that spending show up in your financial controls? Not in your card reconciliation. Not in your AP process. Somewhere else — and "somewhere else" is where control breaks down.

Who is liable when an agent makes a bad purchase? You, because you deployed it? Your vendor, because they built it? The settlement layer, because they facilitated it? We don't have answers yet, and we're not going to get them from the technology community — we're going to get them from the first few expensive lawsuits.

What does "authorization" even mean when the actor isn't human? You can't exactly send a machine a two-factor authentication text. The whole edifice of modern payments is built on the assumption that a person is in the loop somewhere, making a decision. That assumption is now wrong.

What to Actually Do Monday Morning

Here's where I'd start if I were sitting in your chair:

Ask your AI and data vendors how their agents handle external API costs. Not whether they do — whether you have visibility and control. If the answer is vague, that's your red flag.

Map where autonomous spending could be happening in your stack. This isn't just your customer-facing AI chatbot. It's your data pipelines, your ETL jobs, your analytics platforms. Anywhere software is calling external services without a human in the loop.

Start talking to your payment processors and compliance teams about agent authorization frameworks. They probably don't have good answers yet. That's fine. The point is to get the conversation started now, before you're explaining to the board why an agent spent $100K on API calls over a weekend because someone misconfigured a spending limit.

The card networks didn't join x402 because they were bored. They joined because they've watched this movie before — when e-commerce disrupted retail, when mobile disrupted payments, when APIs disrupted software distribution. They know the pattern. The question isn't whether machines will become buyers. They already are. The question is whether you'll have control over what they buy.

But what do I know — I've only watched this cycle play out four times.

---

Want to go deeper? I'm tracking where machine-to-machine settlement is showing up in financial operations and what it means for audit trails, fraud controls, and compliance. If you're seeing this pattern in your organization, I'd like to hear about it: connect with me on LinkedIn or email me directly. This is evolving fast, and the practitioners who figure out the control frameworks early will be in a much better position than the ones playing catch-up in two years.