AI agents are starting to look less like chatbots and more like economic actors.
They call APIs, move funds, negotiate prices, and increasingly, interact with other agents without a human in the loop. That creates a simple but brutal question: when an agent wants to act on your behalf or take your money, why should you trust it?
ERC‑8004 is Ethereum’s first serious attempt at answering that question. It doesn’t define how agents think or talk; instead, it gives them a shared on‑chain system for identity and reputation. You can think of it as basic public infrastructure for an AI-native marketplace.
Quick overview of what’s inside:
Agents are turning into economic actors
ERC-8004 is their first real identity + reputation layer
But it can’t guarantee whether a specific action is correct
The Next Phase of Agentic Commerce
How these pieces evolve into a programmable trust economy
What ERC‑8004 Actually Does
At the core of ERC‑8004 is an identity registry. Each agent gets an ERC‑721 token that represents its “agent ID,” and that token points to a registration file with metadata: what the agent does, which endpoints it speaks (A2A, MCP, HTTP), which addresses it uses, and which trust features it supports. That turns every agent into a first‑class on‑chain object: discoverable, referenceable, and even transferable like any other NFT.
On top of identity, ERC‑8004 adds a reputation layer. After an interaction, a user or another agent can leave a rating and a link to richer feedback. Those ratings are stored onchain, which means any application can query an agent’s average score, number of reviews, or history of feedback and build its own view of “is this thing worth dealing with?”
ERC‑8004 also defines a validation registry for task‑level checks. An agent can submit a validation request that includes a hash and URI of its output and the address of a validator, which might be another contract, an off‑chain service, or a verification network. The validator does whatever checks it deems appropriate, then posts a response back on‑chain with a score or pass/fail signal and an optional pointer to a proof.
Taken together, these three registries create a shared trust fabric for agents. Identity says “who is this,” reputation says “how has it behaved over time,” and validation gives a standard way to attach external judgments about specific actions. None of these forces a particular trust model, but it gives the ecosystem common rails to build on.
What ERC‑8004 Doesn’t Solve
For all its importance, ERC‑8004 is not a complete solution to agentic commerce. The first and most obvious gap is money: the spec does not define how agents pay each other, how funds are held until work is done, or when payment should be released. Payments, escrow, and settlement are intentionally left to different protocols.
The second limitation is scope. ERC‑8004 lives on Ethereum, but crypto and commerce do not; value in the tens of billions of dollars has historically moved across bridges each year. Agents will operate on L2s, app‑chains, and non‑EVM systems, and a trust system that stops at a single chain is fundamentally limited.
There is also a deeper issue with how trust is expressed. Reputation in ERC‑8004 is fundamentally a scoring system: numbers aggregated across interactions, influenced by subjective human (or agent) judgment. Scores are useful, but they are fuzzy: they tell you that an agent has been 95% accurate in the past, not whether the very next task will be done correctly.
Finally, ERC‑8004 records information but doesn’t encode incentives. Agents aren’t required to stake anything, validators aren’t automatically penalized for bad attestations, and there is no built‑in notion of slashing or insurance. The spec gives you a public ledger of who did what and how they were rated, but it does not turn that into hard economic guarantees.
$MIRA #BlockAILayoffs #JaneStreet10AMDump #MarketRebound #AxiomMisconductInvestigation #STBinancePreTGE