AI Agent Discoverability
Lessons from Web & Voice Device Discovery to Enhance AI Agent Accessibility.
AI Agent Discoverability
Lessons from Web & Voice Device Discovery to Enhance AI Agent Accessibility.
The web had a problem with discoverability…websites where out there, but we struggled to find them. So, the notion of a search engine was conceived.
We could enter a search, and based on our search criteria, a list was provided with websites…
This was again the challenge with the advent of Alexa Skills and Google Home Actions…there was no discoverability…
Many skills and actions were developed, people loved the devices but discovery was always the impediment…or rather the lack of discovery.
Also a challenge was the ephemeral nature of voice and the fact that this was all prior LLMs, hence conversations had to be shorter, more command driven etc.
- Also follow me on LinkedIn or on X! 🙂
Search 2.0: Language Model Providers Are Prioritising Distribution to Dominate Next-Gen Search
A Look at How Accessibility Shapes the Future of Search, Data Discovery & Synthesiscobusgreyling.medium.com
Now already LLMs are changing discovery…some call it Search 2.0, others call it Answers, instead of Search (Search to Answers).
We are not only searching, but getting LLMs to search for us, curate information and synthesise the data into a format we want.
Already we see MCP servers which are public facing, performing specific tasks like web search, GitHub search and much more. And these MCP severs can be grouped, searched…hence discovered.
Awesome MCP Servers
A collection of MCP (Model Context Protocol) serversmcpservers.org
Now, companies will have AI Agents, as MCP servers, many-many AI Agents will be public facing.
So there are two aspects here, communication and discoverability.
Communication is inter-AI Agent communication…think of how websites and pages can link to each other.
AI Agents will have to communicate and be discoverable.
The evolution of the web is a good guide to what the AI Agent future looks like in terms of interoperability and discovery.
Domain Name System (DNS)
The Domain Name System (DNS) for websites is a global standard that translates human-readable domain names (for example, google.com) into machine-readable IP addresses, enabling users to locate and connect to websites.
Analogously, the Agent Name Service (ANS) for AI agents provides a directory service that assigns unique, verifiable identities to AI Agents and enables their secure discovery across diverse systems.
It uses DNS-inspired naming and resolution mechanisms to map AI Agent identities to their capabilities and locations, allowing agents to interact in a standardised way.
The ANS draws critical lessons from the web’s early discoverability challenges, where the absence of standardised systems made finding websites cumbersome until DNS and search engines emerged.
Similarly, voice device ecosystems like Alexa Skills and Google Actions have faced discoverability hurdles due to proprietary frameworks that restrict cross-platform compatibility.
ANS addresses these issues by introducing a protocol-agnostic directory with a capability-aware naming system, enabling public-facing AI Agents to be easily located based on their functions, such as natural language processing or task automation.
By incorporating Public Key Infrastructure (PKI) certificates, ANS ensures that AI Agents are not only discoverable but also trusted, mitigating risks like impersonation or malicious interactions that plagued early web ecosystems.
This focus on security and interoperability positions ANS as a foundational infrastructure for the future of AI Agent ecosystems, much like DNS was for the internet.
With potential support from major industry players, ANS could achieve the critical mass needed for widespread adoption.
By aiming for secure discovery, ANS addresses challenges like those seen in platforms such as Alexa Skills and Google Actions, where proprietary systems limit interoperability and discoverability.
As the number of AI Agents grows, ANS’s scalable architecture and lifecycle management mechanisms — such as AI Agent registration and renewal — will ensure that the directory remains current and resilient, paving the way for a robust, interconnected AI agent landscape that mirrors the web’s accessibility and reach.
The graphic shows the architecture and workflow of the Agent Name Service (ANS), a system designed to enable secure discovery and interoperability of AI Agents, drawing inspiration from the Domain Name System (DNS) for the web.
The diagram shows the workflow of how an AI Agent interacts with the ANS ecosystem to achieve secure discovery and communication.
Discovering & verifying AI Agents are standardised.
Here’s a breakdown of each element and the acronyms…
AI Agent
This represents an AI Agent (a user / public-facing AI Agent like a customer service bot) seeking to discover or communicate with another AI Agent.
The AI Agent initiates the process by sending a request for discovery to the ANS Service, similar to how a user queries a website on the web.
ANS Service (Agent Name Service)
ANS stands for Agent Name Service, the core component of this infrastructure.
It acts as a universal directory for AI Agents, akin to DNS for websites.
The ANS Service maintains a registry of AI Agents, mapping their unique identities to their capabilities and locations.
When an AI Agent requests discovery, the ANS Service looks up the requested AI Agent’s information in the Agent Registry and facilitates secure communication.
AI Agent Registry
This is the database or repository where all AI Agents are registered.
It stores critical information about each AI Agent, such as their unique identity, capabilities (natural language processing, task automation) and location.
The Agent Registry works in tandem with the ANS Service to provide lookup services and ensure agents can be discovered efficiently.
Certificate Authority (CA)
The CA (Certificate Authority) is responsible for issuing and verifying digital certificates to ensure the authenticity of AI Agents.
It uses Public Key Infrastructure (PKI) to assign verifiable identities to AI Agents.
When the Agent Registry looks up an AI Agent, it verifies the AI Agent’s identity with the CA to ensure the agent is legitimate and not a malicious entity, addressing security concerns like impersonation.
Registration Authority (RA)
The RA (Registration Authority) validates the registration of AI Agents before they are added to the Agent Registry.
It ensures that only legitimate AI Agents are registered by verifying their credentials and capabilities.
The RA plays a crucial role in maintaining the integrity of the Agent Registry, ensuring that the directory remains trustworthy and up-to-date.
Protocol Adapter Layer
This layer handles the translation of messages between AI Agents that may use different communication protocols (A2A, MCP, ACP, as mentioned in the study).
It ensures interoperability by adapting messages to the appropriate protocol, allowing agents on diverse platforms to communicate seamlessly.
Proposed Workflow
An AI Agent initiates the process by requesting discovery of another agent through the ANS Service.
The ANS Service looks up the requested agent’s information in the Agent Registry.
The Agent Registry verifies the identity of the agent by consulting the Certificate Authority (CA), which uses PKI to confirm authenticity.
Simultaneously, the Registration Authority (RA) validates the agent’s registration to ensure it is legitimate and properly documented in the registry.
Once verified, the Protocol Adapter Layer translates messages between the requesting agent and the discovered agent, enabling secure and interoperable communication.
Relevance to AI Agent Discoverability and Security
This infrastructure addresses the challenges of AI Agent discoverability and security.
The ANS Service and AI Agent Registry provide a DNS-like system for discovering public-facing AI agents, solving the problem of fragmented ecosystems (like those seen in Alexa Skills or Google Actions).
The CA and RA ensure security by verifying identities and maintaining a trusted registry, mitigating risks like impersonation or unauthorised access.
The Protocol Adapter Layer enhances interoperability, ensuring agents can communicate across platforms, much like how the web became accessible through standardised protocols.
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. From Language Models, AI Agents to Agentic Applications, Development Frameworks & Data-Centric Productivity Tools, I share insights and ideas on how these technologies are shaping the future.
COBUS GREYLING
Where AI Meets Language | Language Models, AI Agents, Agentic Applications, Development Frameworks & Data-Centric…www.cobusgreyling.com
Agent Name Service (ANS): A Universal Directory for Secure AI Agent Discovery and Interoperability
The proliferation of AI agents requires robust mechanisms for secure discovery. This paper introduces the Agent Name…arxiv.org
AgentDNS: A Root Domain Naming System for LLM Agents
The rapid evolution of Large Language Model (LLM) agents has highlighted critical challenges in cross-vendor service…arxiv.org