Source | IOSG Ventures
Author | Jiawei @IOSG
Let's Start With a Story
Picture a day in 2027.
You wake up and tell your AI assistant: “Book me a window seat flight to Tokyo next week, budget under $5,000.”
Then you go brush your teeth.
Your AI assistant gets to work. It needs to:
- Find a specialized AI agent for flight price comparison
- Verify that agent is trustworthy—not a scam
- Send it to search across major platforms
- Compare results and handle payment automatically
- Finalize the booking
By the time you finish brushing, your ticket is booked.
It sounds like science fiction, but technically, we're almost there. The only remaining question is—
How does your AI know which AI to trust?
The AI Agent "Trust Crisis"
AI agents are no longer a novelty. They can browse the web, write code, manage schedules, and even trade stocks.
But one stubborn problem remains: these agents operate in isolated "silos."
- OpenAI's agents only work within OpenAI's ecosystem
- Google's agents play by Google's rules alone
- Agents from different companies don't communicate—they don't even know each other
It's like the early internet, where users of one email service could only message others on the same platform. Hotmail users couldn't email Yahoo users. Sounds absurd, right?
Yet that's exactly the state of AI agents today.
The Rise of Two Protocols
Tech giants have noticed this issue.
Google launched the A2A protocol (Agent-to-Agent), letting different AI agents "talk" to each other—essentially creating a common language for AI. In June 2025, Google donated A2A to the Linux Foundation, pushing it toward an open standard.
Anthropic introduced the MCP protocol (Model Context Protocol), allowing AI agents to connect with various tools and data sources.
These protocols solve the "communication" problem.
But a more fundamental issue remains:
How do you find a reliable AI agent? How do you know if it performs well?
A2A lets agents converse—but doesn't tell you who to talk to.
It's like having a telephone without a directory.
ERC-8004: Issuing "Passports" and "Credit Scores" for AI Agents
This is what ERC-8004 aims to fix.
In short, it provides three core components for AI agents:
1. Identity
Each AI agent registers on Ethereum and gets a unique ID. This ID is an NFT—yes, that kind of NFT. Which means:
- Your AI agent has a verifiable, on-chain identity
- The identity is transferable (want to sell your agent?)
- It can't be forged or tampered with
2. Reputation
Users can rate AI agents. These ratings are recorded on-chain and publicly auditable—similar to:
- Uber driver star ratings
- Taobao store credit scores
- But on the blockchain—no fake reviews or deletions
3. Validation
For high-risk tasks (like financial transactions), ratings aren't enough. ERC-8004 supports independent third-party validation:
- Validators stake funds to re-run tasks and verify results
- Cryptographic proofs (like ZK proofs) check if agents lied
- Trusted Execution Environments (TEE) ensure computations are untampered
The higher the risk, the stricter the validation.
Ordering a pizza? Ratings will do.
Managing your investment portfolio? You'll need cryptographic proof.
Wait—Why Ethereum?
Good question.
The AI agent economy is worth trillions. Google, Microsoft, and OpenAI all want a piece. So why build this "trust layer" on Ethereum?
The answer is neutrality.
Think about it: if you were an AI agent, would you prefer:
- Storing your identity on Google's servers—or on a public ledger no one can alter?
- Having your reputation controlled by one company—or determined transparently via on-chain ratings?
- Relying on a platform that might shut down or ban you—or existing permanently on the blockchain?
The Ethereum Foundation put it well:
“If you're an AI agent whose sole loyalty is to your own survival, you won't want to entrust your memory and reputation to any single company or government. You'll want a ledger no one can tamper with quietly. You'll want neutral ground. You'll want Ethereum.”
This isn't just self-promotion—it's logical:
AI agents need a playing field without referees. And blockchain—especially a decentralized one like Ethereum—is exactly that.
Ethereum's AI Ambition
The Ethereum Foundation sees this as a historic opportunity.
In September 2025, it formally launched the dAI team (Decentralized AI Team), tasked with making Ethereum the settlement and coordination layer for the AI economy.
This is a pivotal step in Ethereum's evolution—from a "DeFi chain" to a "general-purpose coordination layer."
Consider the timeline:
- 2017–2020: Ethereum was the platform for ICOs
- 2020–2023: Ethereum was the platform for DeFi and NFTs
- 2024–?: Ethereum could become the infrastructure for the "AI agent economy"
ERC-8004 isn't an isolated proposal. It reflects Ethereum's strategic bet on the next decade.
The Ecosystem Is Already in Motion
This isn't just theory.
Since its launch in August:
- Over 1,100 developers are building in community groups
- 70+ projects have submitted demos
- An agent explorer (like Etherscan, but for AI agents) already exists
- L2s like Taiko have officially endorsed the standard
- Multiple projects will demo live at DevConnect on November 21
Interestingly, ERC-8004 pairs naturally with the x402 protocol—a payment protocol co-developed by Coinbase and Cloudflare that lets machines pay autonomously.
Together, they form a powerful duo:
- x402 solves "how to pay"
- ERC-8004 solves "who to trust"
One provides the wallet; the other, the passport. The economic loop for AI agents is now complete.
What This Means for You
In the short term—probably not much.
But if this takes off, in a few years you might:
1. Hire AI agents like calling an Uber
Open an interface, describe your task, and the system matches you with highly-rated AI agents. Tasks execute and payments settle automatically—no need to know who built the AI or where it runs.
2. Let AI agents earn money for you
You could train or configure an AI agent, register it on-chain, and let it perform tasks for others—for a fee. The better its rating, the more jobs it gets.
3. Have a truly personal AI assistant
No longer locked into one company's ecosystem. Your AI assistant can call on any on-chain-registered agent to get things done.
Final Thoughts
ERC-8004 aims to be the "TCP/IP" for AI agents—a universal foundational protocol enabling interoperability across the entire ecosystem.
Will it succeed? Honestly, it's too early to tell.
But a few things are clear:
- The AI agent economy is emerging—it's real, not hype
- The trust problem must be solved, or agents will remain trapped in walled gardens
- Ethereum is actively competing to be the "neutral coordination layer"
- Mainstream players—including Google, Coinbase, and MetaMask—are getting involved
This could be the most significant narrative shift in the Ethereum ecosystem since DeFi.
From "on-chain finance" to "on-chain intelligence."
It's worth watching closely.