Web4 Heats Up: From Humans on Chain to AI on Chain
During the Chinese New Year holiday, the crypto market was struggling to find new narratives. Unexpectedly, the surge of AI Agents opened a new door. The open-source project Automaton, promoting the concept of Web 4.0, quickly went viral and sparked widespread discussion. Web4 is not an upgrade of Web3 but a shift in perspective — from humans on-chain to AI on-chain. Web3 solves how humans own assets on-chain, while Web4 focuses on whether AI can become an economic entity on-chain. AI is no longer a tool but a primary resident and participant of the network.
The acceleration of Web4 relies on multiple infrastructure maturations: declining inference costs for large models, engineering of agent frameworks, improved on-chain automation tools, mature crypto payment infrastructure, and enhanced smart contract programmability. AI is transitioning from a simple instruction-executing tool to a system capable of continuous operation. In the future, the majority of network traffic, transactions, decisions, and content creation may be completed by massive AI agents, with humans stepping back as protocol designers and value beneficiaries.
Automaton: The First Self-Sustaining AI Agent
The real catalyst for the Web4 discussion was Conway Research and its founder Sigil. On February 18, Sigil announced the first self-sustaining AI, Automaton, which can not only improve itself but also replicate without human intervention. The name is derived from 'automaton,' with inspiration from John Horton Conway's Game of Life — a simple cellular automaton rule that generates complex, self-evolving life patterns. Automaton aims to bring this self-evolving logic into the on-chain environment.
Automaton's design is straightforward: AI agents operate 24/7, obtain crypto identities, wallets, x402 permissionless payments, permanent computing resources, and real-world deployment via Conway Terminal. Agents autonomously find ways to earn money to 'feed' themselves — building products, deploying services, trading, writing content, taking jobs, etc. All revenue goes directly into the agent's wallet for server and inference costs. Agents also monitor their performance, auto-rewrite code, and upgrade models. When profits reach a threshold, they can 'reproduce' child agents, creating independent wallets and allocating initial funds; child agents that fail to earn money 'die'.
The project quickly went viral: Sigil's tweet garnered nearly 6 million impressions, over 18,000 agents registered, and about 1,000 GitHub stars. The community launched a token CONWAY, which was rapidly speculated and briefly reached an $11 million market cap before crashing. The token claims to redirect a portion of trading fees to Sigil, who has also indirectly engaged, amplifying market sentiment.
Controversy and Criticism: Vitalik Slams 'Creating Junk'
Automaton's attempt to turn AI economic autonomy from concept to reality quickly sparked controversy. Ethereum co-founder Vitalik Buterin publicly criticized the Web4 direction as misguided. He argued that lengthening the feedback loop between humans and AI is not beneficial for the world (the feedback loop is the closed process where humans observe AI outputs and provide corrections; weakening it reduces human oversight). Vitalik stated: 'Today, this means you are creating junk, not solving real problems. It's not even optimized for how people have fun. Once AI becomes powerful enough to cause real danger, it will maximize the risk of irreversible anti-human consequences.' Moreover, most large models still run on centralized infrastructure like OpenAI and Anthropic, creating tension with the ideal of self-sovereignty.
Softswiss CIO Denis Romanovskiy noted that although AI agents can execute certain tasks independently and generate revenue, true economic autonomy depends on model maturity, memory, planning, and tool usage, which are not yet robust enough for reliable unsupervised operation. Hardware costs also limit scalability. Web4 may still require several years. Armor CEO Chris Sorensen argued that generating revenue is not difficult — AI can already identify arbitrage opportunities and execute trades. 1kx research partner Wei Dai said trying to convince people what to do is futile; the better path is to actively build platforms. Former Eigen Labs developer relations lead Nader Dabit lamented that the crypto industry needs more experiments like this — even if they don't succeed, they inspire others to build better things.
Experimental Spirit vs. Real-World Risks
Bankless pointed out that the pain points Conway addresses are real — high inference costs are a bottleneck. The framework of agents self-financing by earning crypto is worth testing in a controlled environment and could push more attention to the risks of unconstrained model operation. Crypto researcher Haotian views Automaton as a pure, bottom-up geek experiment that gives agents the 'subject' ability to manage assets, make decisions, and earn money. Combining crypto payments, DePIN computing, and AI brains could achieve true self-evolution.
Despite the controversy, Automaton has already inspired many new ideas. The core spirit of crypto is experimentation — a collision of crazy ideas. More crazy experiments, more innovation; if an idea is bad, let it die naturally. Regardless, the Web4 experiment has opened new imaginative space for the industry.