Original: Deep Tide TechFlow

On January 28, Ethereum’s official account announced that the ERC-8004 protocol is about to launch on the mainnet.

We already mentioned this in an article last October. If you’re completely unfamiliar with it, you can refer here: 《x402 Is gradually getting more and more cutthroat—digging into the new asset opportunities in ERC-8004》

Actually, it has an official name: “Trustless Agents,” sending credentials to trusted-less agents. Put into plain human language, it’s roughly:

Issue on-chain ID to AI Agents.

The Ethereum Foundation rarely puts this much effort behind pushing an ERC standard. They specifically set up a team called dAI, listed ERC-8004 in the 2026 strategic roadmap, pulled in Google, Coinbase, and MetaMask to draft it together, and in November they even held a Trustless Agents Day at DevConnect to drum up momentum.

The last time Ethereum pushed a standard this seriously was with ERC-20 and ERC-721.

One defines a token, and one defines an NFT.

Now it’s AI’s turn?

Ethereum’s AI anxiety

Why the rush?

Take a look at some data. According to Cookie.fun’s statistics on the market-cap distribution of AI Agent tokens, the two chains Solana and Base together capture 96%. There are only a handful of AI Agent projects on Ethereum mainnet that you can actually name.

There are only a handful of AI Agent projects on Ethereum mainnet that you can actually name.

In April 2025, ETH’s exchange rate against BTC fell to 0.017, the lowest in five years. Back then, everyone said Ethereum wasn’t the future.

When DeFi was hot, Ethereum was the home turf. When NFTs were hot, Ethereum was still the home turf. When AI Agents heated up, the home turf changed hands.

Solana processes 36 million transactions per day, while Ethereum mainnet processes 1.13 million. With high gas fees and slow speed, developers cast their votes with their feet. When Virtuals Protocol launched on Base, earlier ai16z had chosen Solana, and even Coinbase’s own AI project wasn’t placed on Ethereum mainnet.

Ethereum needs a new story.

ERC-8004 might be the opening of that story.

Let’s review ERC-8004 again

Back to the standard itself.

So how exactly does ERC-8004 work for issuing on-chain ID to AI Agents?

You don’t need to understand any technical details. You just need to know there are three ledgers.

The first is the identity ledger. Based on ERC-721, each AI Agent mints an NFT to prove, “I am me.”

The second is the reputation ledger. It records the Agent’s past performance—who has used it, what kind of evaluations it has received, and whether it has done anything bad.

The third is the verification ledger. It lets third-party institutions stamp and endorse the Agent—such as “this Agent has passed some security audit.”

Together, these three ledgers solve one problem: when two AI Agents meet on-chain, how do you know whether the other one is reliable?

Before, the answer was: you can’t know—you can only rely on people. The answer from ERC-8004 is: check the on-chain records.

This set of ideas wasn’t invented by Ethereum itself.

Its underlying logic comes from Google’s A2A protocol released last year—Agent-to-Agent—so AI can talk to each other and call each other. On top of that, ERC-8004 adds a layer:

Trust backed by the blockchain.

Google’s A2A solves the communication problem; Ethereum’s ERC-8004 solves the trust problem. One handles the talking, one handles proving identity.

Is issuing IDs a good business?

Let’s make a bold guess. Ethereum’s logic might be something like this:

For an AI Agent to be truly useful, it needs to be able to manage money on its own. Not tweeting, not chatting—directly operating on-chain assets. Signing transactions, calling contracts, cross-protocol arbitrage…

Right now, nobody dares to do this at scale. The reason is simple: how do you know that this Agent won’t just transfer your money away? The ClawdBot that’s been exploding in popularity these days has already had community users share related negative incidents.

Web2’s solution is platform endorsement. If you use OpenAI’s API, the trust comes from OpenAI. If something goes wrong, you go find OpenAI.

Web3 doesn’t have that. Agents are open source; deployment is permissionless; they run on-chain with nobody overseeing them. You call a service from a stranger Agent—who’s behind it, whether the code has issues, and whether it has a history of doing evil… you can’t find any of that.

To put it bluntly, ERC-8004 essentially copies the traditional finance KYC process onto the blockchain. And Ethereum is betting that once AI Agents start dealing with real money, this will become a must-have.

For DeFi protocols to integrate external Agents, they first need to check the Agent’s on-chain identity. For institutions to use Agents for transaction execution, they first need to examine its historical records. Audit firms can issue on-chain certifications for Agents—just like performing security audits for smart contracts.

This is a strategic foothold in competition.

Ethereum knows it has already lost on the execution layer, but nobody has occupied the trust layer yet. Institutional recognition, the security audit ecosystem, and TVL scale—these are Ethereum’s existing assets. ERC-8004 packages these assets into a standard, rushing to define what “AI Agent compliance” looks like before others do.

The question is: does this need exist right now?

Standards come before demand

After laying out Ethereum’s playbook, it’s time to face reality. What are on-chain AI Agents doing right now?

After the AI meme wave last year ran its course, and given how much the big AI companies’ AI products advanced by leaps and bounds over the past one or two years, there aren’t many people still paying attention to on-chain AI Agents.

But they’ve still made progress.

For example, ai16z has already rebranded to ElizaOS, evolving from a single Agent into a cross-chain platform. Virtuals Protocol is building an AI DAPP and plans to move into physical robots in 2026. Also, other AI Agents like the ones in Surf can automatically execute DeFi trading strategies.

But then the issue arises: do they really need ERC-8004?

Luna users trust Luna because it’s built by Virtuals’ core team. Agents on ElizaOS are used because they run within the ElizaOS framework. Surf helps you execute strategies, and a lot of the time, it’s because you trust the application itself.

Trust comes from the platform, not from on-chain identity.

The scenario imagined by ERC-8004 is: a strange Agent comes to find you. There’s no platform endorsement, no brand awareness—you can only judge whether it’s reliable by looking at on-chain records.

When would this scenario happen?

When AI Agents truly achieve autonomous calling across protocols, platforms, and organizational boundaries. An Agent borrows money from Aave, trades on Uniswap, then makes yield on another protocol—through the entire process, without human approval…

But this scenario doesn’t exist right now.

For now, even if today’s AI Agents are complex in function, in essence they still operate within a single platform. They don’t need to prove themselves to unfamiliar protocols because they simply won’t go knock on the doors of unknown protocols.

Given the current heat in the crypto market, they also have no reason to knock on each other’s doors—unless they can work together to create a new narrative.

So ERC-8004 solves a problem that may arise in the future.

If AI Agents evolve from toys into tools, then Ethereum’s trust infrastructure becomes valuable. If the scale of the Agent economy becomes large enough, and cross-platform calling becomes the norm, ERC-8004 can collect tolls.

There are a lot of “if”s.

So, for this future-facing setup, the first movers are likely to be institutions.

By late 2025, SharpLink Gaming announced it would allocate $170 million into Ethereum re-staking strategies. Around the same time, exchanges saw net outflows of more than 23k ETH, flowing into private wallets and staking protocols.

This money might be buying Ethereum that’s 12 to 18 months ahead.

For retail investors, ERC-8004 isn’t really a great catalyst.

Bet on ERC-8004 itself? It’s an open standard with no token, so you can’t invest directly—you can only look for some associated small projects. Betting on Ethereum is also not impossible, but Ethereum’s price is influenced by too many factors, and AI Agents are only one of the narratives.

So, right now there’s no clean target that lets you precisely bet on the thesis that “AI Agents need on-chain identity.”

Ethereum isn’t entirely the infrastructure for AI, and Ethereum’s identity anxiety won’t be fully alleviated just because AI becomes fully realized. Doing the business of AI IDs is still a long and arduous road.

Welcome to join ChainCatcher’s official community

Telegram subscription: https://t.me/chaincatcher;

Official Twitter account: https://x.com/ChainCatcher__