For ten years this industry has argued about how to bring a billion people into crypto. The conversation has often been centered on better wallets, simpler onboarding, gasless transactions, education campaigns. Entire conferences were built around the question. I attended many of them. The answer, when it finally arrived, came from a direction almost nobody was expecting. The first billion users of blockchain infrastructure will be machines.

The evidence is no longer speculative. The x402 protocol, released by Coinbase in May 2025 to let AI agents pay for services over plain HTTP, crossed 100 million cumulative agentic transactions on Base alone within three quarters of going live, according to Chainalysis. In July 2026 the Linux Foundation formally launched the x402 Foundation with 40 member organizations. Look at who took governing seats: Visa, Mastercard, American Express, Stripe, AWS, Google, Circle, Shopify. All three major card networks joined a standards body for machine payments settled in stablecoins. When the incumbents of the payment world sit down at the same table as their supposed disruptors, the debate about whether something is real has ended.

A More Honest Customer Than We Deserve

Here is what strikes me most about this shift, after four decades in technology. The machine turns out to be a more honest customer than the human ever was.

A human user can be persuaded. Marketing can convince a person to tolerate a slow network, to forgive an outage, to stay loyal to a chain because of a community, a mascot, a memory of the 2021 bull run. Entire ecosystems have survived for years on sentiment alone. All of that means nothing to an autonomous agent. It evaluates finality time, fee structure and reliability, selects the best option and executes. The whole decision takes milliseconds, and sentiment plays no part in it. If your network is slower or more expensive than the alternative, the agent leaves before you knew it arrived.

The early data already shows this to be true. Chainalysis found that the wallets driving x402 activity behave nothing like typical retail users. Conversion from cautious first tests to active external payments improved fourfold in six months, and the bulk of transferred value has shifted into transactions of a dollar or more. These are not tourists collecting airdrops. They are software systems funding operational wallets because the workflow demands it.

This makes the agentic economy something genuinely new. It is the first market in the history of this industry where infrastructure competes purely on technical characteristics such as throughput, latency, cost and uptime, while brand strength counts for nothing. For those of us who came from engineering rather than promotion, this is the market we always wanted to operate in.

Designing for Users Without Hands

What does a network built for this kind of customer actually require? Consider what an agent lacks. It has no hands to click through wallet pop-ups, no eyes to read a warning banner, no patience to wait through congestion, and no intuition to sense that something feels off. Every assumption baked into today's user experience, the confirmation screen, the seed phrase ritual, the "please wait" spinner, is an assumption that there is a human being on the other side. Remove the human and most of that scaffolding becomes dead weight.

An agent-ready network needs sub-second finality, because a machine negotiating a chain of dependent payments cannot tolerate settlement uncertainty. It needs fees measured in fractions of a cent, because agentic commerce lives on micropayments. It needs parallel execution, because a million agents do not queue politely. And it needs account abstraction at the protocol level, so that programmable wallets, spending policies and delegated permissions are native features instead of fragile add-ons.

I will be candid about our own position here. When our team designed Venom's asynchronous, sharded architecture with account abstraction built into the base layer, nobody was talking about AI agents. We were designing for a different demanding customer, the payment infrastructure of governments and large institutions, workloads where volume is enormous, latency intolerable and failure unacceptable. It turned out that a central bank settlement system and a swarm of autonomous agents ask for almost exactly the same things. I will not claim we predicted the agentic economy. We built for requirements that the agentic economy now happens to share. Sometimes engineering for the hardest customer prepares you for the customer you never expected.

The Metrics Have Expired

One more consequence deserves attention. Every adoption metric this industry has worshipped since 2017 assumed the user was a person. Active addresses, a number our own research team publicly challenged this year as inflated by Sybil farms and bots, became meaningless when it became possible for one agent to operate thousands of wallets and one wallet to serve thousands of agents. Daily active users, social followers, community size, all of it measured human attention. Machines pay no attention. They pay invoices.

So the scoreboard has changed. The networks that win the next decade will be judged the way we judge power grids and clearing systems, on volume settled, on failures avoided, on cost per operation.

Which leads to a conclusion I find almost poetic. The race to onboard the next billion users will be won by a network that a billion never sees. Its users cannot read headlines. They will just keep routing around everything that fails to work, at machine speed, forever. Build accordingly.