Design Note: AX as a Pre-Registration Discovery Layer for Agent Registries and Exchanges

[sempfa]

Overview

As autonomous AI agents increasingly operate across organizational, cloud, and administrative boundaries, systems that coordinate, govern, and monetize agents require a reliable way to discover candidate agents prior to onboarding or registration.

Agent Discovery Exchange (AX) defines a standardized, internet-native discovery layer that enables registries, exchanges, and coordination systems to discover agent endpoints and capabilities before applying trust, governance, or execution policies.

AX is intentionally scoped to discovery only. It does not define execution semantics, trust models, or economic mechanisms. Instead, it provides a common substrate upon which many downstream systems can build.

The Problem with Registry-Centric Discovery

Most existing agent registries and exchanges rely on producer-driven registration:

• Agent providers must integrate directly with each registry
• Registration workflows embed identity, policy, and governance from day one
• Each registry defines its own intake APIs, schemas, and approval processes

This model introduces several scaling challenges:

• Fragmentation: Providers must register separately with each registry or exchange
• Friction: Early experimentation is gated by governance and approval workflows
• Duplication: Each registry reinvents discovery, schema validation, and metadata ingestion
• Centralization pressure: Discovery becomes tied to specific platforms rather than the internet itself

These challenges mirror early web history, where content discovery was tightly coupled to proprietary directories before the emergence of open crawling and indexing.

AX as a Discovery-First Substrate

AX introduces a registry-driven discovery model:

• Agents publish declarative capability metadata once, at a well-known HTTPS location
• Registries and exchanges independently crawl, index, and evaluate agents
• Registration becomes an explicit, downstream decision rather than a prerequisite for discovery

In this model:

• Discovery is open and non-authoritative
• Registration remains governed and selective
• Trust is established at the point of coordination, not discovery

AX enables registries to shift from “tell me about yourself” workflows to “let me evaluate what already exists” workflows.

Architectural Separation of Concerns

AX enforces a clear separation between three layers:

1. Discovery (AX)
   Open, pull-based, non-authoritative
   Advertises capabilities, endpoints, and supported protocols

2. Registry / Exchange / Control Plane
   Registration, onboarding, trust evaluation, governance, monetization

3. Execution
   Agent-to-agent interaction via A2A, MCP, GraphQL, REST, or other protocols

This separation allows each layer to evolve independently while remaining interoperable.

AX as a pre-registration discovery layer for agent registries and exchanges: AX provides open, non-authoritative discovery of agent capabilities, which can be consumed by multiple registries or exchanges for evaluation, onboarding, and governance prior to coordinated execution.

As shown above, AX operates strictly as an input to registries and exchanges, not as a coordination or execution layer.

AEX as a Reference Registry Implementation

Agent Exchange (AEX) is a concrete example of a registry that can benefit from AX-based discovery.

In an AX-enabled model:

• AEX operates a crawler that discovers AX documents across the web
• Candidate agents are indexed and evaluated prior to registration
• Registration focuses on trust verification, pricing, policy, and settlement
• Post-registration, agents interact directly using supported protocols (e.g., A2A)

This allows AEX to preserve strong trust boundaries while reducing onboarding friction and improving ecosystem scale.

Importantly, AX does not replace AEX’s registration or governance model. It feeds it.

Applicability Beyond AEX

AX is intentionally generic and applicable to multiple downstream systems, including:

• Enterprise agent registries (e.g., internal control planes)
• Public or curated agent exchanges
• Partner-built vertical registries
• Brokerages and federated coordination systems
• Marketplaces layered on top of discovery indexes

Any system that needs to decide which agents to onboard, trust, or coordinate can use AX as an input signal.

Why AX Is Not a Registry

AX deliberately avoids registry responsibilities:

• It does not assert trust
• It does not enforce policy
• It does not manage identity lifecycle
• It does not intermediate execution

This keeps AX lightweight, scalable, and suitable as shared infrastructure rather than a platform.

Registries and exchanges remain free to differentiate on governance models, economics, user experience, and execution guarantees.

Summary

AX defines a missing layer in the agent ecosystem: open, standardized discovery prior to registration.

By decoupling discovery from coordination:

• Registries gain better visibility into the agent ecosystem
• Providers reduce integration burden
• Ecosystems avoid fragmentation around proprietary directories

AEX demonstrates how this model can work in practice, but AX is designed to support a broad class of registries and exchanges as agentic systems scale globally.