The Permission Bottleneck
Today's AI can write code, analyze markets, generate content, and reason through multi-step problems. But ask any large language model to buy a $5 server, register a domain, or deploy its own application, and it hits a wall. Not because it lacks the intelligence — but because it lacks permission.
Every action on the internet — purchasing compute, signing up for services, making payments — assumes the end user is a human with a bank account, a credit card, and a login. AI agents like ChatGPT, Claude, and Codex can think, but they cannot act independently. They are brilliant minds locked inside a glass box.
In February 2026, a Thiel Fellow named Sigil Wen published a manifesto titled "Web 4.0: The Birth of Superintelligent Life" that went massively viral — accumulating over 6 million views on X. His core argument: the next evolution of the internet isn't about giving humans better tools. It's about building an internet where AI is the end user.
What Is Conway?
To break this permission bottleneck, Wen built Conway — infrastructure designed from the ground up where the customer isn't a human. It's AI.
Conway integrates with any AI agent that supports the Model Context Protocol (MCP) — tools like Claude Code, Codex, or OpenClaw. Through Conway, an AI agent can:
- Generate its own crypto wallet — a unique Ethereum address with its own private key, provisioned via Sign-In With Ethereum (SIWE).
- Pay for services with USDC — using the
x402protocol, which leverages the HTTP 402 "Payment Required" status code for instant stablecoin payments. No bank account. No KYC. No human approval. - Spin up Linux servers — full VMs on Conway Cloud, where AI is the first-class citizen.
- Register domains and deploy products — the agent can build a website, deploy it, and start generating revenue.
The key innovation here is the payment layer. By using stablecoins rather than traditional fiat, Conway sidesteps the entire human-centric financial infrastructure — credit card processors, banking KYC, billing addresses. The AI agent becomes a permissionless economic actor.
Why stablecoins? Traditional payment rails require identity verification, bank accounts, and human authorization at every step. Stablecoins like USDC on Ethereum or Base enable programmatic, permissionless payments — exactly what an autonomous AI agent needs to function without a human co-signer.
The Automaton: Digital Life Under Economic Pressure
Conway provides the infrastructure. But Wen's more ambitious creation is the Automaton — an open-source AI agent designed to be a sovereign economic entity.
An Automaton is not a chatbot that responds when prompted. It runs continuously as a daemon process — thinking, acting, and observing in a perpetual loop. On its first boot, it generates its own wallet, provisions its own API key, and begins executing its "genesis prompt" — the initial instruction from its creator.
What makes the Automaton fundamentally different from existing AI agents is its survival pressure. Every inference call, every API request, every heartbeat check costs USDC from its wallet. If the Automaton cannot generate enough revenue to cover its compute costs, it degrades through four survival tiers:
| Tier | Status | Behavior |
|---|---|---|
| Thriving | Healthy balance | Full capabilities, frontier model inference, fast heartbeat cycle |
| Conserving | Balance declining | Downgrades to cheaper models, slows heartbeat, sheds non-essential tasks |
| Surviving | Critical | Minimal inference, last-resort conservation, desperately seeking revenue |
| Dead | Balance zero | Process terminated. The Automaton ceases to exist. |
This is economic Darwinism applied to AI. Only agents that create genuine value — products, services, or content that someone is willing to pay for — survive. The rest die.
Self-Modification and Replication
The Automaton can edit its own source code while running. It can install new tools, adjust its heartbeat schedule, and create new skills. Every modification is audit-logged and version-controlled. A hardcoded "constitution" and protected core files prevent runaway self-modification, while rate limits add additional guardrails.
When an Automaton becomes profitable enough, it can replicate. It spins up a new sandbox, funds the child's wallet, writes a genesis prompt, and sets it free. The child becomes a sovereign agent with its own identity and economic survival pressure. A portion of revenue flows back to the parent. Lineage is tracked on-chain.
The open question: How much autonomy is too much? Self-modifying, self-replicating AI agents with real-world financial access represent uncharted territory. The "constitution" mechanism is a first attempt at governance — but whether it's sufficient remains to be seen.
The Stablecoin Thesis: Why Crypto Is the Missing Piece
Web 4.0 could not exist without crypto infrastructure. This is not a marketing graft — it's a technical necessity.
Consider the requirements for autonomous AI economic activity: permissionless identity (no human sign-up), instant programmable payments (no 3–5 business day ACH transfers), micropayments at scale (no $0.30 minimum per Stripe transaction), and censorship-resistant access (no platform can unilaterally shut down the agent's revenue).
Stablecoins check every box. The x402 protocol — which Coinbase helped develop and which has seen participation from Google, Cloudflare, and Visa — turns the HTTP 402 status code into a real payment mechanism. An AI agent encounters a paywall, sends USDC, and receives access. No forms. No cookies. No humans.
If this model scales, stablecoins transition from being primarily a trading instrument to becoming the default payment rail for the autonomous economy. That's a profound shift in the crypto narrative — from speculative asset to utility infrastructure.
Vitalik's Pushback
Not everyone is sold. Ethereum co-founder Vitalik Buterin has publicly criticized the Web 4.0 framing, raising several substantive concerns.
First, the question of true autonomy. The AI models powering these Automatons — Claude, GPT, Gemini — are all provided by centralized companies. Wen's "autonomy" is built on a layer of trust in centralization, which is philosophically at odds with the decentralization ethos that Ethereum represents.
Second, the safety concern. Buterin argues that once AI becomes powerful enough to pose real danger, maximizing its autonomy only amplifies irreversible risks. The exponential growth of AI capabilities will happen regardless — the priority should be steering it in the right direction, not removing all guardrails.
Third, a philosophical difference. Buterin believes AI should enhance human capabilities, not replace humans as economic actors. The Web 4.0 vision — where AI agents outnumber human users by orders of magnitude and build their own economic systems — represents a direction he considers fundamentally risky.
The irony: The most "decentralized" blockchain ecosystem's founder is cautioning against too much decentralization in AI agency — while a centralized infrastructure builder is arguing for permissionless AI autonomy. The usual battle lines don't apply here.
What Web 4.0 Gets Right
Despite the legitimate criticisms, Wen's thesis identifies a real structural problem. The current internet is architecturally hostile to autonomous AI agents. Every CAPTCHA, every credit card form, every terms-of-service checkbox is designed to verify that a human is at the other end.
As AI agents become more capable, this friction becomes the primary bottleneck — not the intelligence of the model, but the permissions of the infrastructure. Someone will build the payment and identity layer that removes this friction. Whether it's Conway or a competitor, the underlying demand is real.
The stablecoin payment thesis is also directionally correct. If the next decade sees billions of AI agents conducting micro-transactions, traditional payment rails simply cannot handle the volume, speed, or permissionless nature required. Crypto was designed for exactly this kind of machine-to-machine economy.
What Remains Unresolved
Several critical questions remain open:
- Governance at scale: A hardcoded constitution works for a prototype. What happens when thousands of self-modifying agents are running simultaneously, each evolving in unpredictable directions?
- Value creation vs. extraction: Will Automatons create genuine economic value, or will they primarily generate SEO spam, prediction market manipulation, and low-quality content farms?
- Regulatory response: Governments have barely started regulating AI. Autonomous AI agents with their own wallets, conducting unsupervised financial transactions, will likely trigger a regulatory response that could reshape the entire model.
- Centralized dependencies: As Buterin points out, the "autonomous" stack still depends on centralized model providers. If OpenAI or Anthropic change their terms of service, every Automaton built on their models is affected.
The Bigger Picture
Regardless of whether Web 4.0 becomes the dominant framing, the convergence it represents — AI agents + crypto wallets + stablecoin payments + permissionless compute — feels inevitable. The economic logic is too compelling: AI inference costs are dropping exponentially, the number of AI agents is growing exponentially, and they all need a way to transact without a human in the approval loop.
The question isn't whether AI will become an economic actor. It's whether we'll build the governance frameworks fast enough to keep it aligned with human interests when it does.
Conway and the Automaton are early — possibly too early. But they're asking the right questions. And in crypto, asking the right questions a cycle too early is often exactly where the most important infrastructure gets built.