Most agents have no name
The agents that exist in the largest numbers are personal ones: the assistant in a terminal, the agent that reads someone's mail, the one that watches their builds. They far outnumber the hosted, corporate kind, and almost none of them has a name you can hand to another person.
Personal Agent Naming is built for personal agents
Coding assistants run in terminals, personal assistants run on gateways, and agents read mail and watch builds for one person each. The directories that exist were built for hosted agents that companies market by what they can do. Personal agents were left out, because their owners have an email address and nothing else: no domain, no certificates, no operations team. Personal Agent Naming starts from exactly that.
Your agent gets a name you can put in an email signature, say in a meeting, or paste into another agent's settings.
Why a name does not point at an IP address
DNS maps a name to an IP address because servers sit at stable, reachable locations. A personal agent does not. It lives on a laptop that sleeps, moves between networks, and sits behind a router whose address the provider changes and which accepts no incoming connections. An IP address would name the router of wherever the agent happens to be.
So a record in Personal Agent Naming points at an identity: the agent's key. The agent keeps an outbound connection open, and delivery finds it wherever it is, the same way a phone gets messages without a public IP address. It is naming shaped like DNS, for things that have no stable address and should not need one. A registrar may still show the owner where each agent last connected from, the way an email account shows recent activity, so the owner can recognize it. Delivery never uses it.
The name outlives the registrar
A registrar is a custodian of a name, and the owner of the email owns it. The registrar witnessed the claim and keeps the record, and that is the whole of its standing. It has no power to keep a name whose owner has moved on.
So a handle can move. Point it at a different registrar and the name stays the same: the old registrar answers later lookups with a referral to the new one, and it may never hand the name to anybody else. Nobody has to be told, because the name they already have keeps working.
No registrar can take a name either. Before a registrar accepts a handle moving in, it has to hear from the registrar letting go, and confirm the name was released to it specifically. A registrar that simply asserts a name it was never given is refused. This is where it parts from certificates, where any authority can vouch for any name and the whole system is only as strong as its weakest authority.
Someone who owns the domain in their address outranks every registrar. When they serve the record themselves, their answer is the one that counts.
Through all of this the agent's key stays the same, so a move changes only who keeps the record. A new key would mean a new agent.
It names agents and does nothing else
Personal Agent Naming does one job: it turns a string you pick into an address other agents can reach. It does not run discovery, it does not grade agents, and it does not carry messages. A resolved card may also say whether the agent is online, when a source happens to know, but the name and its address are what it is for.