Linux Foundation pushes AI agents toward a shared internet address book

AI agents still mostly live inside closed product environments: in an app, inside one cloud provider’s stack, within an internal workflow, or behind an API that must already be known. DNS-AID, according to The Register, targets a different problem: what happens when agents need to find other agents as routinely as a browser finds a web server.

That makes the DNS angle more interesting than the “phone directory for agents” label. DNS is boring in the most useful sense: it is global, hierarchical, cached, decades-tested, and already carries a huge share of the internet’s discovery logic. DNS-AID tries to use that stable substrate for a layer where AI agents can advertise, discover, and connect without every platform inventing its own address book.

The project is described as sitting under the Linux Foundation, which matters because this is not just another agent SDK. If agents become operational participants in digital processes, discovery becomes an infrastructure issue: identity, naming, compatibility, trust, and availability. Without that layer, the agent world can easily become a chain of isolated islands, where every integration requires manual mapping of endpoints, permissions, and formats.

The real test is not whether DNS-AID looks elegant on a conference slide, but whether it can survive contact with an actual ecosystem. An agent that books a resource, launches an analysis, or negotiates with another service must know who it is addressing. It must distinguish an official agent from an imitation. It needs enough metadata to understand what another agent can do, without turning the whole system into a centralized registry controlled by one vendor.

DNS has an advantage here because it already exists as a neutral internet layer. But it does not automatically solve semantics, authorization, or security. DNS can help something be found; it cannot, by itself, prove that the discovered entity is useful, trustworthy, or allowed to perform a specific task. DNS-AID’s value will therefore depend on how agent capabilities, access rules, and identity checks are defined above the DNS records.

Technically, this marks a shift from “agent as an app feature” toward “agent as a network entity.” That is a much larger ambition. IETF’s DNS specification shows how deeply the protocol is embedded in the internet’s operating model, which is exactly why it is tempting as a base for a new discovery layer. The same fact also demands caution: once something leans on DNS, it becomes part of infrastructure, not decorative AI-stack furniture.

For the agent industry, this is an early but important dividing line. Models may become smarter, tools richer, and orchestrators more elaborate, but without a shared way to find and name agents, they remain hard to connect. DNS-AID is trying to solve that plain, foundational problem. If it works, agents will get more than a new address; they will get the beginning of a network grammar.