If you are building an AI agent system today, you are likely operating on borrowed time. While protocols like MCP and A2A have successfully standardized how agents talk to one another, the infrastructure required to find those agents in the first place is currently a chaotic, uncoordinated sprawl.
A new survey from an Ericsson team — including former IETF Chair Jari Arkko — brings this reality into sharp focus. Their July 2026 Internet-Draft maps the current state of agent discovery across the IETF, AAIF, 3GPP, and various open-source projects. The authors categorize the landscape by evaluating how different mechanisms handle five core components: lookup, name, description, trust, and interaction.
The sheer volume of competing standards is staggering. The survey catalogs over 15 distinct discovery mechanisms. Nowhere is this friction more apparent than at the DNS level, where four separate proposals — DNS-AID, DN-ANR, AID, and AgentDNS — are currently vying for dominance.
The survey highlights eight critical gaps that prevent true interoperability. The most significant is the absence of a standard for semantic discovery. Other missing pieces include the lack of interoperable federation, fragmented metadata schemas, and a total absence of standardized lifecycle management for agents.
The unresolved challenge of trust bootstrapping adds another layer of complexity. DNS lookup alone is insufficient; while TLS can provide a domain certificate, it does not verify the identity of the agent behind that domain. We are now seeing a competition for control, as DNS-based discovery mechanisms clash with emerging on-chain agent registries.
For technical builders, this landscape survey serves as a necessary reality check. The actual interoperability battle will not be won at the protocol level, where MCP and A2A already hold ground, but at the discovery layer. Until a standard emerges, agent systems will remain trapped in isolated pockets.