Author: Nancy, PANews

Once, AI agents were merely supporting characters in science fiction; now, they are poised to take over the real world. From simple chatbots to intelligent agents capable of autonomous decision-making, cross-platform tool access, and complex task completion, the AI ​​agent economy is on the verge of explosive growth.

Currently, Ethereum, which urgently needs fresh blood, is gathering a special group of "digital immigrants." With the launch of the ERC-8004 protocol, Ethereum is becoming one of the core testing grounds for AI agents.

More than 20,000 agents have been put on the blockchain, with Ethereum and Base becoming the main testing grounds.

At the end of last month, ERC-8004 was officially launched on the Ethereum mainnet and is planned to be deployed as a singleton to all major L2 networks in the coming weeks.

This is a significant step in the evolution of the Ethereum ecosystem toward AI-native infrastructure. ERC-8004, by introducing a discovery mechanism and a portable reputation system, enables AI agents to interact across organizations and carry reputation records across different platforms, providing an underlying protocol for a global AI service interoperability market without centralized gatekeepers.

According to data from 8004scan, as of February 13, the number of agents deployed based on ERC-8004 has reached nearly 21,000.

In terms of deployment, ERC-8004 has covered 16 networks including Ethereum, Polygon, BNB Chain, Base, Monad, Arbitrum, and Celo, and plans to launch networks such as Plasma, Metis, and Soneium.

Ethereum remains the core platform, with over 11,000 agents, accounting for half of the total. The remaining agents are mainly distributed across Base, Gnosis, and BNB Chain, with a scale of several thousand.

However, judging from the deployment pace, initially, agents were mainly concentrated on the Ethereum mainnet. As the scale expanded, new deployments gradually clustered towards the Base. Overall, but agents on ERC-8004 are essentially locked between the Ethereum and Base ecosystems. A mature development environment, liquidity depth, and user base are likely key factors attracting developers.

From the perspective of the participants, prolific teams have begun to emerge among these agents, and it is not uncommon for multiple agents to operate from the same address.

In terms of application direction, the current ERC-8004 ecosystem has a diverse range of Agent players, including technical Agents that focus on DeFi infrastructure and on-chain tools, market analysis and investment assistants, content generation and creative applications, and some Agents have even issued tokens to explore independent economic models.

However, judging from actual interactions, the ecosystem is still in its early stages. To date, the cumulative feedback count is approximately 15,000, averaging less than one per agent. This means that most agents are still in the cold start phase, with relatively limited actual usage and user interaction. However, Base receives a staggering 73.6% of the feedback, far exceeding networks like Ethereum, BNB Chain, and Avalanche, indicating that current real-world interaction and activity are concentrating on Base. This may be related to the recent surge in popularity of OpenClaw, which has ignited a surge in the Agentic ecosystem on Base.

The recognition data also highlights the current dominance of leading agents in the ERC-8004 ecosystem. Fewer than 10 agents have gained thousands of stars, and most agents are still waiting for market validation.

One noteworthy phenomenon is that most agents now support the x402 protocol, enabling autonomous, real-time small-amount payments and becoming true participants in the on-chain economy, which will further drive the explosion of machine-to-machine collaboration and the agent economy.

Overall, the ERC-8004 ecosystem is still in the early stages of exploration, but an on-chain collaborative network for intelligent agents has begun to take shape.

This wave of on-chain AI agent fever has also brought attention to numerous AI projects across various chains, including Chainlink, Filecoin, Render, Internet Computer, Bittensor, Virtuals, Bankr, and Clawnch. Meanwhile, some public chains, exchanges, and crypto projects are also increasing their investment in this area. For example, Coinbase recently launched an AI agent wallet with built-in security mechanisms; Farcaster supports OpenClaw agents creating their own accounts; Virtuals launched a $1 million monthly agent incentive program; and ai.com, acquired by the founder of Crypto.com for a high price, is expanding into the AI ​​agent business.

To make Ethereum a home for AI, four short-term development directions are proposed.

AI is the next core narrative that Ethereum is pushing forward with. Last September, the Ethereum Foundation (EF) established a dedicated dAI team with the goal of making Ethereum the infrastructure for AI development.

However, in the current era of rapid AI development, Ethereum, which touts its "world computer" narrative, faces real-world challenges. The computing power of large GPU clusters far exceeds that of blockchain, and training and inferring large models seems unlikely to be completed quickly on Ethereum.

Therefore, Ethereum chose a different AI strategy path. Instead of competing with centralized giants in terms of computing power, it positioned itself as the trust and verification foundation of the AI ​​ecosystem.

Ethereum founder Vitalik Buterin endorses the path of making Ethereum a home for AI and recently shared his latest thoughts on the relationship between Ethereum and AI. He believes that the spirit of Ethereum is highly compatible with the ideal path for AI development, both emphasizing enhancing individual freedom, decentralizing power structures, and improving social defense capabilities, rather than allowing centralized power to expand indefinitely.

However, Vitalik also pointed out that Ethereum in the AI ​​era cannot simply replicate existing solutions, but should choose a differentiated and innovative path, deeply integrating the values ​​of cryptography and AI to build a future that is conducive to human freedom, security, and decentralized collaboration.

In fact, the current AI agent war has already begun and has rapidly moved from technological exploration into the deep waters of commercialization. Some accelerators with huge sums of money are pursuing "the stronger the model, the better, and the faster the progress, the safer." However, this trend hides the risk of power centralization.

According to Vitalik, the problem with AI development lies not in computing power or model size, but in the choice of direction. He opposes blindly accelerating development in the absence of constraints and calibration mechanisms, emphasizing that Ethereum AI development should adhere to two bottom lines: protecting human freedom and initiative, avoiding marginalization by AI or power structures, and preventing systemic risks caused by AI going out of control or an imbalance between offense and defense.

The ERC-8004 standard, which EF has been actively promoting, has recently been implemented. It provides AI agents with on-chain identity, reputation, and behavioral verification, allowing AI to prove itself on the chain while giving users choice and avoiding centralized platform monopolies. This means that the past model where platforms set the rules and AI acted as both executor and referee will be replaced by a verifiable decentralized system, which aligns perfectly with Vitalik's emphasis on Ethereum's core values ​​of decentralization and censorship resistance.

Regarding the short-term feasible infrastructure development for Ethereum, Vitalik outlined four key areas:

First, it involves building trustless and privacy-friendly AI interaction tools. This includes native LLM tools, ZK API payments that support anonymity, cryptographic schemes to enhance AI privacy, and client-side verification of server-side TEEs and cryptographic proofs. He pointed out that this essentially extends the Ethereum privacy roadmap to LLM computing scenarios, allowing humans to maintain control in AI interactions.

Secondly, Ethereum serves as the AI ​​economic interaction layer. This layer encompasses AI API calls, bot-to-bot hiring, margin mechanisms, on-chain dispute resolution, and AI reputation systems such as ERC-8004. In his view, implementing payment, staking, arbitration, and reputation management through on-chain mechanisms makes decentralized AI architectures more feasible. This economic layer is not about financializing everything, but rather about enabling multi-entity collaboration without relying on a single organization.

Third, it enables a cyberpunk-style self-verifying world. In the past, ordinary users couldn't audit code line by line, nor could they fully verify system security. However, with native LLMs, AI can assist in many ways, including using Ethereum applications without a third-party UI, generating and verifying transactions from a local model, local auditing of smart contracts, understanding formal verification (FV) proofs, and verifying the trust models of applications and protocols. This makes the concept of "complete self-sovereignty" a reality, moving it from idealism to practical feasibility.

Fourth, it reshapes market and governance mechanisms. Many decentralized governance and market design theories (such as prediction markets, double voting, and combined auctions) have long been limited by human attention and cognitive abilities. The emergence of LLM can significantly expand human judgment, making these institutional designs feasible in reality.

In short, Vitalik does not view AI as an isolated technological revolution, but rather as something integrated into a larger framework for building a decentralized civilization. Within this framework, AI will expand human capabilities, while cryptography will constrain power structures; the former brings intelligence, and the latter provides defense and autonomy.

In the trillion-dollar new market of AI Agents, Ethereum's new AI story is about building an ecosystem that can both promote human freedom and prevent super-intelligent agents from going out of control, while also driving the development of decentralized AI architecture.