This blog post is based on a paper titled “Discovering Agents for Discovery: The Case for DNS,” authored by Ramachandra Rao Seethiraju, Sameer Thakar, Karthik Shyamsunder, and Eric Osterweil. This blog was co-authored by all four authors.

For more than three decades, domain names have served as a foundational identity layer for internet applications. Initially used to identify early network services such as TELNET, FTP, and email, they later became essential to web browsing and a growing range of online services. Their enduring value lies in their ability to provide unique, stable, neutral, and widely recognized identifiers across changing technologies and use cases.

Now, the Domain Name System (DNS) has the potential to serve as a foundational element to certain key aspects of artificial intelligence (AI) discovery, as we discuss in a new Internet Engineering Task Force (IETF) Internet-Draft.

Intersection of the DNS and AI Agent Communication

The internet’s infrastructure and services are beginning to rely on AI, and AI agent deployments are now growing at an unprecedented rate. As they become more prolific across the internet, AI agents are increasingly operating across organizational and technological boundaries. While many AI agents still operate within relatively “closed world” environments, trusted and secure cross-platform interaction is poised to become a foundational challenge.

In this new AI-enabled world, AI agents have some basic operational requirements. They need to discover one another, identify each other’s capabilities, establish trust, authenticate their endpoints, and exchange information securely across heterogeneous environments. At the internet’s scale, these requirements introduce important challenges around discoverability, interoperability, integrity, and transaction efficiency. This frames large design and deployment challenges.

Fortunately, the DNS already provides a globally deployed and proven naming and resolution infrastructure for the internet that is well suited to these challenges. Combined with DNSSEC and DNS-based Authentication of Named Entities (DANE), the DNS also provides mechanisms for integrity protections, authenticated associations, and trust establishment. These characteristics become more relevant as AI agents continue expanding beyond platform-specific ecosystems.

Phases of AI Agent Communication

Foundationally, AI agent communication can be deconstructed into two logical phases: AI agent discovery and runtime interaction. We’ve developed an evaluation framework to propose a basic set of metadata that may be necessary and sufficient for conducting AI agent discovery. Our results indicate that this metadata set fits well within practical DNS operational limits under common deployment conditions. This enables the initial discovery and bootstrapping using existing DNS transaction models and infrastructure already deployed at internet scale. Our research shows that the “necessary and sufficient” metadata set consists of a protocol identifier, coarse-grained capabilities, a service endpoint, and identity verification. As agentic AI continues to mature, additional sets of runtime metadata may continue to evolve, but the set we have identified will likely remain necessary and sufficient for discovery.

Binding AI Agent Metadata to Domain Names

Building off our research, our recent IETF Internet-Draft DNS-Based Agent Naming (DAN): AIDISCA and AIINDEX Resource Records for AI Agent Discovery structures agentic AI discovery as a use case for DANE associations—i.e., in a similar fashion to the way DANE enables authenticated associations for Transport Layer Security (TLS). This protocol is designed to enable AI agents to locate and connect to each other using necessary metadata and secure DNS associations. It proposes two new DNS Resource Record (RR) types:

• AIDISCA, a DANE-like domain name-to-AI agent association record, and
• AIINDEX, a DNS-based discovery mechanism for enumerating AI agent publication points within a domain name.

AIDISCA provides an atomic representation of the metadata necessary to establish connectivity with an AI agent, including protocol identifiers, capabilities, service endpoints, and certificate associations. Correlating this interoperable bootstrap metadata directly with domain names reduces external dependencies and minimizes additional resolution steps during discovery.

AIINDEX enables domain-scoped discovery of AI agent publication points without requiring extensive DNS crawling or other discovery systems.

Together, these mechanisms allow AI agents to retrieve authoritative metadata through a minimal number of DNS transactions, reducing lookup overhead, improving consistency, and enabling low-latency interaction establishment across organizational and platform boundaries.

Binding agent metadata to domain names provides organizations an opportunity to leverage the trust they have built over the years with their domain names.

Publication of Royalty-Free Licensing Terms

We have already started working with the community on extending existing internet naming and trust infrastructure toward AI agent discovery, and we will continue to socialize, gather feedback, and fine-tune the various proposals, including these new DNS resource record types. Additionally, as part of this work, we announced a royalty-free public license to support standardization of DAN (see the DAN Internet-Draft IPR declarations for the official language).

We are encouraged by these timely efforts toward scalable, interoperable, and authoritative interaction across organizational and technological boundaries.