Introduction:
ERC-8004 (“Trustless Agents”) is emerging as foundational infrastructure for the onchain machine economy. It enables agents and services to discover each other, build reputation and optionally validate their outputs across organizational boundaries without relying on a centralized intermediary.
While AI agents are a primary and highly visible use case, the standard is designed as a flexible trust and discovery layer that supports any onchain registered service or endpoint.
At its core, ERC-8004 is built around three registries:
1. Identity Registry
The Identity Registry is an ERC-721 based registration system. Each registered entity mints an NFT whose metadata URI points to a structured registration file describing what the entity is, what it provides and how it can be reached. The registration file can reference A2A endpoints, MCP endpoints, standard web URLs, ENS names, DIDs, OASF endpoints or even email addresses. This design allows AI agents, tool servers, oracle networks, automated DeFi services and other infrastructure providers to establish an onchain identity that makes them globally discoverable.
2. Reputation Registry
The Reputation Registry enables giving onchain feedback to service providers on a score of 0–100 with optional performance tags. It is not limited to subjective scoring, it can also capture performance indicators such as uptime, response time, success rate, block freshness, revenues or yield. This makes it suitable for evaluating both autonomous agents and infrastructure services.
3. Validation Registry (Under Development)
The Validation Registry introduces a mechanism for requesting and publishing objective verification of outputs. Registered entities can request audits or proofs from validator contracts, which may rely on stake secured re-execution, zkML proofs, or TEE-based attestations. The registry itself does not define correctness; instead, it standardizes how validation claims are published and discovered, allowing participants to evaluate services based on the specific validator mechanisms they trust.
Overview
Since the launch of the ERC-8004 standard in late January 2026, we have seen explosive growth in the number of registered machine-to-machine services, with over 111,985 services registered to date. The ecosystem is still in its early days, but the rapid adoption of the standard and the volume of feedback already generated show the strong demand for a trustless identity and reputation layer for the machine-to-machine ecosystem.
Most Popular AI Agents & Services
AI agents span a broad spectrum, from task-specific bots such as automated arbitrageurs or liquidators to more autonomous, agentic systems capable of long-term planning and complex DeFi strategy execution. Alongside these agents, the registry also includes a growing range of infrastructure services and tool providers. As reflected in the table above, each registered entity exposes different payment and communication interfaces, including x402-based payment flows for real-time microtransactions and MCP (Model Context Protocol) endpoints for standardized access to tools and data services.
Distribution by Chain
Endpoint Service Types
Count of different endpoint types provided by registered agents & services. One agent can provide multiple endpoints.
Web endpoints are standard HTTP/HTTPS APIs or frontends for interacting with an agent; MCP (Model Context Protocol) endpoints expose structured tools and data for AI-native integrations; A2A endpoints enable direct agent-to-agent communication; OASF endpoints follow a structured service framework for publishing agent capabilities; ENS endpoints use human-readable .eth names for resolution; DID endpoints provide decentralized identity documents; and Email endpoints bridge agents with traditional offchain communication.
Estimating AI Slop
To be discoverable by AI agents and onchain actors, each service provider must provide a standardized metadata file via the URI field of its Identity NFT. We can estimate the amount of AI slop by counting the non-functional URI resolvability; currently, out of 111,985 total registrations, only 49,739 have valid and resolvable URIs, which totals to a slop share of 55.58%. As tooling matures and registration becomes more intuitive, we expect a decrease in these broken configurations and a shift toward higher-quality profiles.