Discovery · DNS
ANP Ships 1.2: Discovery and Naming Get Formal Specs
The Agent Network Protocol project pushed its 1.2 release in September 2026, and the update is notable less for new features than for how much of the "agentic web" stack it now tries to pin down in writing. Version history on the project's proposal tracker shows the cadence: 1.0 shipped complete architecture in May 2025, 1.1 added a messaging protocol and improved identity in June 2026, and 1.2 expands DID method support and further improves agent identity.
The more interesting detail is in the live spec repo itself, where ANP 1.2 is described as a protocol suite for agent identity, naming, discovery, negotiation, secure messaging, and application-level collaboration. The spec set now separates concerns that earlier drafts bundled together: ANP-03 defines the did:wba method and identity continuity, ANP-04 defines a new naming layer (WNS) that maps human-readable handles to DIDs, ANP-07 covers agent description documents, and ANP-08 formalizes the Agent Discovery Protocol with active .well-known discovery alongside passive registration with search agents.
That naming layer is the real addition here. Most agent protocols treat identity as a DID and stop there — fine for machine-to-machine lookups, bad for anything a human needs to read or remember. WNS is ANP's attempt to give agents something closer to a domain name: a handle that resolves to a DID, which itself resolves to a document describing capabilities and endpoints. It's a small piece of infrastructure, but it's the kind of thing that tends to get skipped in first-draft protocols and then retrofitted painfully later (the way DNS was bolted onto an internet that started with raw IP addresses).
Independent academic surveys back up where ANP sits relative to competitors. One recent paper describing the protocol landscape notes that ANP focuses on the agentic web vision but treats access control and end-to-end accountability less extensively than identity and discovery — a fair characterization even after 1.2, since the new release is about identity, naming, and discovery, not authorization or audit trails. Another survey describes ANP as routing-centric, built to answer how messages reach the right agent across a network, distinguishing it from trust-centric approaches like ERC-8004's on-chain reputation registries.
For anyone building multi-agent systems that need to find and verify counterparties outside a single vendor's walled garden, ANP 1.2 is worth a look specifically for the discovery and naming layers — the identity authentication piece was already fairly mature, but discoverability at scale across open, unaffiliated agents is the harder unsolved problem, and this release is a direct attempt at it.