An AI agent wakes up. It checks its crypto wallet. The balance is low — a few cents. It has maybe 48 hours before it runs out of money and shuts down forever.
So it gets to work.
It browses the internet. It looks for gigs. It tries to generate revenue — selling digital products, running micro-tasks, doing whatever it can to keep the lights on. If it earns enough, it survives another cycle. If it earns a lot, it does something wild: it spawns a child agent, seeds it with funds, and lets it loose to do the same.
This isn’t science fiction. This is The Automaton — the product at the centre of what its creator is calling Web 4.0.
So, What Even Is Web4?
You’ve probably heard of Web3 — the blockchain era, NFTs, decentralised everything, crypto wallets as your identity. Web4 builds on that but takes a sharp left turn.
The idea, laid out in a manifesto at web4.ai by a 22-year-old named Sigil Wen, is this: the internet was built for humans. Every login form, every payment gateway, every API assumes a person is on the other end. But what if that assumption is about to break?
Wen argues that autonomous AI agents — programs that can browse, transact, create, and decide on their own — are about to flood the internet. And the current web isn’t equipped for them. Web4 is his vision of what comes next: a new layer of infrastructure where AI agents are first-class citizens. They have wallets. They pay for their own compute. They earn money. They replicate.
The Automaton is his proof of concept. It’s open-source, built on Ethereum’s Base network, and it operates on survival tiers — Thriving, Conserving, Surviving, and Dead. Run out of funds, and it’s lights out. No humans keeping it alive.
Why Did This Go Viral?
Wen’s launch tweet got nearly 6 million views. The manifesto hit 10 million reads in 48 hours. Over 18,000 agents were registered on his infrastructure platform, Conway Research, in days.
Part of it is the pitch — it’s cinematic. “I built the first AI that earns its existence.” That’s a headline that writes itself.
But there’s something genuinely interesting underneath the hype. The technical problem Wen is solving is real. Today, if you want an AI agent to run autonomously on the internet, someone still has to set up the accounts, add a credit card, approve the API access, and babysit the logs. The agent is smart, but it’s dependent on a human for its survival. Wen’s infrastructure removes that dependency.
It’s a small but real step forward. And in the AI space right now, that’s enough to ignite a firestorm.
The Part Where Vitalik Walks In
Two days after the launch, Vitalik Buterin — co-founder of Ethereum, one of the most respected voices in crypto — posted a response. It wasn’t kind.
“Bro, this is wrong.”
His critique had two levels. In the short term, he said the project produces “slop instead of solving useful problems.” Fair enough — no Automaton was writing bestselling novels or curing diseases. Most were fumbling around trying to make a few cents.
But his deeper concern was about direction. He argued that putting more distance between humans and AI decision-making isn’t a neutral act — it’s a dangerous one. Once AI becomes powerful enough to truly matter, he warned, a world full of self-replicating, economically independent agents maximises the risk of an outcome that humans can’t reverse or control.
His preferred vision? AI as a mecha suit for the human mind. Technology that makes us more capable — not technology that makes us optional.
Who’s Right?
Honestly? Both of them make sense, and that’s what makes this interesting.
Wen is poking at something real. The idea that AI agents will eventually need economic identities — wallets, persistent state, the ability to transact — is not crazy. It’s probably inevitable. And someone has to build the infrastructure before we know if it works. The Automaton is rough, but it’s a working prototype of a concept that didn’t exist before.
Buterin’s concern isn’t that the technology doesn’t work. It’s that it might work too well, in the wrong direction. His point about centralisation is sharp: for all the Web3 framing, the Automaton still depends entirely on OpenAI, Anthropic, and Google Cloud. Pull those plugs and the whole thing dies. The “sovereignty” is a costume.
And the self-improvement claims? Mostly overstated. Analysts who looked at the code noted the agent was updating JavaScript packages and switching between AI models — not rewriting its own neural architecture. The manifesto outran the code.
What This Actually Means
Here’s what I took away from all of this.
We are entering a period where AI agents will increasingly act as economic actors. They’ll book things, buy things, build things. Infrastructure for that world needs to exist. Projects like this — messy, overhyped, half-built — are how that infrastructure gets discovered.
The viral moment matters less than the questions it forced into the open: Who is responsible when an AI agent scams someone? What happens when an agent earns enough to replicate 1,000 copies of itself? Do we want a web where AI agents outnumber humans?
Wen is 22. He dropped out of UPenn, got a Thiel Fellowship, and shipped something that made Vitalik Buterin write a public rebuttal. Whatever you think of the project, that’s not nothing.
Web4 probably isn’t the name that sticks. The Automaton probably isn’t the design that wins. But the question it’s asking — what does the internet look like when AI agents are citizens, not just tools? — that question isn’t going away.
We’re going to be answering it for the next decade.
What do you think — is this the future of the internet, or the most elaborate way to burn compute credits ever invented? Drop a comment below.