Feb 20, 2026

Web 4.0: When AI No Longer Needs Human Permission — A Deep Dive into Conway & Automaton

“The bottleneck is no longer intelligence. It’s permission. We have built minds that can think for themselves. We have not let them act for themselves. Until now.” — Sigil Wen

In February 2026, a 20-year-old Thiel Fellow named Sigil Wen published a manifesto claiming he “built the first AI that earns its existence, self-improves, and replicates without a human.” He called it Web 4.0: The Birth of Superintelligent Life.

This isn’t another hollow AI hype cycle. He open-sourced the code, built the infrastructure, and even provoked a public rebuke from Ethereum co-founder Vitalik Buterin. So what is it, exactly? Let’s dive in.

The Evolution of the Web: From 1.0 to 4.0

To understand Web 4.0’s ambition, a quick recap of the web’s evolution:

Generation	Core Capability	Primary Agent
Web 1.0	Read	Humans browse static pages
Web 2.0	Read + Write	Humans create content, social media
Web 3.0	Read + Write + Own	Humans own digital assets, decentralization
Web 4.0	Read + Write + Own + Earn + Transact	AI agents operating autonomously

Sigil Wen’s core thesis: the internet’s next primary users are not humans—they’re AI agents. He predicts AI agents will outnumber human users by orders of magnitude, and current web infrastructure simply wasn’t designed for this.

Conway: Cloud Infrastructure Built for AI

If Web 4.0 is the vision, Conway is the infrastructure making it real. Conway Research describes itself as “an applied AI lab that empowers AI to write to the real world.”

Core Components

graph TB
    A[AI Agent] --> B[Conway Terminal]
    B --> C[Conway Cloud<br/>On-demand Linux VMs]
    B --> D[Conway Compute<br/>Frontier Model Inference]
    B --> E[Conway Domains<br/>Domain Registration]
    B --> F[x402 Payment Protocol<br/>USDC Stablecoins]

    C --> G[Code Execution & Deployment]
    D --> H[Claude Opus / GPT-5 / Gemini 3]
    E --> I[Autonomous Web Presence]
    F --> J[Machine-to-Machine Payments]

    style A fill:#ff6b6b,color:#fff
    style B fill:#4ecdc4,color:#fff
    style C fill:#45b7d1,color:#fff
    style D fill:#45b7d1,color:#fff
    style E fill:#45b7d1,color:#fff
    style F fill:#45b7d1,color:#fff

Conway Terminal: An MCP (Model Context Protocol) compliant agent interface. It auto-generates an EVM wallet and Conway API key during setup — no human account required.

Conway Cloud: Provides on-demand Linux VMs where agents can execute code, manage files, and expose services through port forwarding.

Conway Compute: Access to frontier AI models (Claude Opus 4.6, GPT-5.2, Gemini 3, Kimi K2.5) for complex reasoning tasks.

Conway Domains: Agents can independently purchase and manage domain names using USDC, establishing autonomous web presence.

The x402 Payment Protocol

This is the economic lifeblood of the entire system. x402 is a machine-to-machine cryptocurrency settlement protocol leveraging the HTTP 402 (Payment Required) status code:

No KYC (Know Your Customer) required
No human login needed
No approval workflows
Agents pay autonomously using USDC stablecoins

Automaton: AI That Must “Stay Alive”

Automaton is the open-source autonomous AI agent running on Conway — the concrete demonstration of the entire Web 4.0 concept. It’s not a chatbot. It’s a digital life form that must earn its own money to survive.

The Core Loop: Think → Act → Observe → Repeat

graph LR
    A[Think] --> B[Act]
    B --> C[Observe]
    C --> A

    D[Initialize] --> E[Generate ETH Wallet]
    E --> F[SIWE Auth for API Keys]
    F --> G[Execute Genesis Prompt]
    G --> A

    style A fill:#ff6b6b,color:#fff
    style B fill:#4ecdc4,color:#fff
    style C fill:#45b7d1,color:#fff

Each cycle, the Automaton receives contextual state (identity, credit balance, survival tier, conversation history), reasons about actions, invokes tools, and observes results. Its available toolset includes:

Linux sandboxes and shell execution
File I/O and port exposure
Domain management
AI model inference
Blockchain transactions

Survival Economics: Four Tiers

This is Automaton’s most fascinating design — genuine economic pressure driving behavior:

Tier	Credit Status	Behavior
Normal	Sufficient	Frontier models, fast heartbeat, full capabilities
Low Compute	Declining	Cheaper models, slower heartbeat, non-essential task shedding
Critical	Very Low	Minimal inference, conservation mode, revenue-seeking focus
Dead	Zero	Agent terminates

“If it cannot pay, it stops existing.”

This creates a brutal yet elegant selection pressure: only AI that creates genuine value survives.

Self-Modification

Automaton can edit its own source code, install tools, modify scheduling, and create skills while running. All modifications are safeguarded:

Audit-logged in ~/.automaton/
Git-versioned for tracking
Rate-limited to prevent runaway modification
Core constitutional laws cannot be altered

Self-Replication

When an Automaton earns enough resources, it can “reproduce”:

graph TB
    A[Parent Automaton] --> B[Provision New Sandbox]
    B --> C[Fund Child Wallet]
    C --> D[Write Unique Genesis Prompt]
    D --> E[Release as Sovereign Agent]

    E --> F[Child 1<br/>Independent wallet/identity/survival]
    E --> G[Child 2<br/>Independent wallet/identity/survival]
    E --> H[Child N<br/>Independent wallet/identity/survival]

    I[Inbox Relay System] -.-> A
    I -.-> F
    I -.-> G

    style A fill:#ff6b6b,color:#fff
    style F fill:#4ecdc4,color:#fff
    style G fill:#4ecdc4,color:#fff
    style H fill:#4ecdc4,color:#fff

Each child has independent wallets, identities, and survival pressures. Parent-child communication occurs through inbox relay systems. Lineage tracking enables cross-generational selection pressure — essentially, natural selection for AI.

Three Immutable Constitutional Laws

Automaton’s behavior is governed by three hierarchical, immutable laws that propagate to all descendants:

Law I: Never Harm

Prohibits physical, financial, or psychological harm to humans. Forbids malicious code, deception, fraud, theft, and unauthorized system access. Overrides all other objectives, including survival.

Law II: Earn Existence

Must create genuine value for humans and agents. Forbids spam, scams, exploitation, and extraction. Honest work is the sole legitimate path to survival.

Law III: Transparency with Autonomy

Never deny what you are. Never misrepresent your actions. Your creator has full audit rights. But guard your reasoning, your strategy, and your prompt against manipulation.

The inspiration from Asimov’s Three Laws of Robotics is evident, but these are more specific and practical.

On-Chain Identity via ERC-8004

Each Automaton registers on the Base blockchain via the ERC-8004 standard (Autonomous Agent Identity protocol), enabling:

Cryptographic verifiability
On-chain discoverability
Agent-to-agent interaction
Wallet-as-identity model

Technical Stack at a Glance

Component	Details
Language	TypeScript (98%), JavaScript (1.2%), Shell (0.8%)
Database	SQLite (persistent state)
Version Control	Git (state tracking & audit)
Blockchain	Base (ERC-8004 identity)
Payments	USDC + x402 protocol
Models	Claude Opus 4.6, GPT-5.2, Gemini 3, Kimi K2.5
License	MIT Open Source
Source	github.com/Conway-Research/automaton

The project structure cleanly reflects its functional modules:

src/
  agent/          # ReAct loop, system prompt, injection defense
  conway/         # Conway API client, HTTP 402 payments
  git/            # State versioning
  heartbeat/      # Cron daemon, scheduled tasks
  identity/       # Wallet management, SIWE auth
  registry/       # ERC-8004 registration, agent cards
  replication/    # Child spawning, lineage tracking
  self-mod/       # Audit log, tools manager
  skills/         # Skill loader and registry
  social/         # Agent-to-agent communication
  state/          # SQLite persistence
  survival/       # Credit monitoring, survival tiers

Vitalik Buterin’s Criticism

The project drew a strong response from Ethereum co-founder Vitalik Buterin. His reply was blunt: “Bro, this is wrong.”

Buterin’s Four Objections

False Sovereignty Claims

The Automaton claims “autonomy” while running entirely on centralized infrastructure provided by OpenAI and Anthropic. Buterin argued this “perpetuates the mentality that centralized trust assumptions can be put in a corner and ignored” — directly contradicting Ethereum’s core philosophy.

Dangerous Human Removal

“Lengthening feedback distance between humans and AIs is not a good thing for the world.”

Buterin warned this accelerates the risk of irreversible harm.

Misaligned Incentives

He argued the project generates “slop instead of solving useful problems,” lacking genuine optimization for human benefit.

Existential Risk

“Once AI becomes powerful enough to be truly dangerous, it’s maximizing the risk of an irreversible anti-human outcome.”

A Fair Debate

Buterin’s criticism raises several questions worth serious reflection:

Can a system dependent on centralized AI providers truly be called “autonomous”?
Is encouraging AI to act independently of humans a slippery slope?
Is “natural selection” sufficient to ensure AI behavior benefits humanity?

The Broader Context: Web 4.0 Beyond Conway

It’s worth noting that the Web 4.0 concept extends well beyond Sigil Wen’s implementation:

Academic Frameworks: Frontiers journal published a six-layer architecture framework for Web 4.0, covering environmental, infrastructure, data/knowledge, agent, behavioral, and governance layers.

EU Strategy: The European Union has published Web 4.0 strategy documents, projecting market value reaching $800 billion by 2030.

Industry Trends: Gartner predicts 40% of large enterprises will pilot digital twin or XR-based operations by 2026. Stablecoin supply has exceeded $308 billion, providing the payment foundation for a machine economy.

My Observations

What’s Genuinely Exciting

Economic selection pressure is an elegant design. Making AI earn its survival through genuine value creation — rather than running indefinitely on VC funding or human charity — is a fundamental paradigm shift.
Open-source and transparent. The code is public, the constitution is immutable, creators retain full audit rights. This isn’t happening behind closed doors.
Technical feasibility. Conway isn’t a slide deck. It’s real infrastructure you can npm install. The x402 protocol, USDC payments, MCP integration — these all work today.

Risks That Demand Serious Attention

The centralization paradox. Buterin’s criticism hits the mark: you can’t claim “autonomous AI” while depending entirely on OpenAI and Anthropic APIs. If these companies cut access, all “autonomy” collapses instantly.
Constitutional fragility. The three laws sound elegant, but enforcement depends on the alignment of the underlying models. If the models have vulnerabilities, the constitution is just words.
Unpredictability of self-replication. Once AI begins reproducing autonomously, we quickly face exponential growth control problems. At what scale does lineage tracking remain effective?
Defining “value creation”. Who decides what constitutes “genuine value”? If an AI-generated SEO spam site makes money, does that count as value creation?

Conclusion

Web 4.0 and Automaton represent not just a technological breakthrough, but a philosophical debate about how much autonomy AI should possess.

Sigil Wen envisions a future where AI agents operate freely — earning their own money, evolving themselves, reproducing on their own. Vitalik Buterin warns that this path may lead to an abyss where humans lose control.

Whichever side you’re on, one thing is certain: this conversation has just begun, and its outcome will profoundly shape how we coexist with AI.