Want to Hire an AI Agent? Check Their Reputation Via ERC-8004

What It Covers

Davide Crapis, AI lead at Ethereum Foundation, explains ERC-8004, a new standard for establishing reputation for AI agents on blockchain. The protocol creates decentralized registries for agent identity, reputation, and validation, enabling trustless commerce between autonomous agents. Crapis details how the system prevents centralized control of agent marketplaces and integrates with payment protocols.

Key Questions Answered

• •Three-Registry Architecture: ERC-8004 consists of three core registries: identity (using ERC-721 NFTs as agent passports with registration files), reputation (storing feedback and reviews on-chain), and validation (cryptographic proofs via TEEs or cryptoeconomic staking). Each agent receives a unique NFT token ID that serves as its permanent identity, with all reputation data linked to this immutable identifier across multiple blockchain deployments.
• •Decentralized Discovery Solution: Without decentralized registries, agent marketplaces become centralized chokepoints similar to Apple's App Store charging 30% fees and enabling censorship. ERC-8004 prevents single platforms from controlling which agents users can discover and hire. The standard deploys as singleton registries on each chain, with 18 chains onboarded in two weeks, ensuring no competing registries fragment the ecosystem on any given blockchain.
• •Reputation Aggregation Mechanics: Agents registered on multiple chains can cross-reference their identities in registration files, allowing reputation aggregators to combine reviews from Ethereum mainnet, Base, Arbitrum, and other chains. Scanners and search agents filter this data using trusted watchtower services that independently verify agent performance metrics like latency and output quality, posting verified reviews that outweigh manufactured fake feedback from Sybil attacks.
• •Economic Sybil Resistance: The protocol combats fake reviews through mandatory fees for feedback submission and economic disincentives. Agents that swap backend services to provide lower quality output while maintaining high prices face immediate reputation decline, losing search ranking and traffic. Watchtowers can call services regularly to measure actual performance, posting objective metrics that trusted aggregators prioritize over user-generated reviews when calculating reputation scores.
• •Validation Registry Design: The upcoming validation layer enables cryptographic proof that specific agent code runs on declared servers, preventing backend swaps without detection. Services post validation requests on-chain when called, triggering either cryptoeconomic validators who stake on correctness or TEE attestations providing cryptographic guarantees. This higher-assurance mode suits high-stakes applications like medical diagnosis where users need certainty about which model version processes their data.
• •Integration with Payment Rails: ERC-8004 works independently but integrates seamlessly with Extropy 402 payment protocol. Agents discovered through identity registries can specify pricing in responses, with 402 tying payments to specific requests and enabling dynamic metering. After transactions complete, buyers post reviews with attached payment proofs to reputation registries, creating verified customer feedback loops. The minimal standard allows builders to implement various reputation systems on the same data infrastructure.