An agent is about to become software's most common user: a program that reads a task, calls five other programs, and hands back a result. For that to work at any scale, one agent has to find another agent and trust that it is who it says it is. That problem has a name, and nobody owns it yet.

The registry problem

When you want to reach a website, you type a name and DNS resolves it. DNS is a shared layer run by thousands of operators under common rules, and it works because no single company controls it. Agents have no equivalent. A tool that needs a specific capability cannot look it up in a shared directory. It has to be wired to a vendor's catalog, and that vendor decides what exists.

Identity is the harder half

An agent acting for a law firm, a hospital, or a teenager with a budget needs to prove three things: who it represents, what it is allowed to do, and that its last answer was not rewritten in transit. Today those proofs live inside platform accounts. An OpenAI agent is an OpenAI object. An Anthropic agent is an Anthropic object. The credential does not travel across the boundary, so an agent cannot carry its identity from one provider to another. It is born in a silo and dies there.

Discovery has the same shape

Every major provider is building a directory of agents: who built them, what they do, how to call them. Each directory is real, useful, and closed. A buyer can browse one store, not the whole market. An agent cannot advertise itself once and be found everywhere, the way a domain name can. The result is fragmentation, and fragmentation is a tax on everyone except the gatekeeper.

One tension decides the outcome

A registry is only valuable if it is complete, and it is only complete if it is neutral. A neutral registry makes no money, because the whole point is that nobody takes a cut. So the companies with the resources to build the registry are the companies least likely to build a neutral one. They will build walled catalogs and call them registries, and for a while that will be enough.

History says the walled version fails

Email worked because addresses were portable across servers. The web worked because DNS and HTTP were open standards anyone could implement. Instant messaging stayed fragmented for a decade, and every attempt to unify it through one vendor collapsed under its own tollbooth. The pattern is consistent: a layer that names and finds things has to be public to scale, and public layers are built by standards bodies, foundations, and the occasional indifferent giant, not by the vendor that stands to profit from owning the door.

A registry is only valuable if it is complete, and it is only complete if it is neutral.

That is the position our industry is in. The people building agents are busy making them smarter. The people who could build the registry are busy building moats around their own catalogs. In the gap sits a piece of infrastructure that will decide who gets to participate in the agent economy: the agent that can be found and verified across every platform, or the agent that only exists inside one.

What a working registry needs

A working registry needs three unglamorous things.

  • A portable identifier. Something like a domain name for agents, that survives a move between providers.
  • A verifiable claim layer. So an agent can prove who it represents and what it is authorized to do without revealing more than it must.
  • A discovery index. Anyone can publish to and query it, run under shared rules rather than a single operator.

None of this requires new science. Every piece already works somewhere: DNS for naming, TLS certificates for claims, email and the web for open transport. What is missing is the incentive. A neutral registry is a public good, and public goods are boring, unowned, and essential. The firms that build them rarely get rich. The firms that skip them, and the firms that try to own them, are the ones whose agents eventually cannot talk to anyone else's.

The workshop's stake

We automate the middle for a living, and a registry is the middle of the middle. We would rather build on a neutral layer than be the ones holding the door, because a business built on owning the door is one standards decision away from being gone. The agents are coming. The question is whether they will meet in a public square or a shopping mall, and the answer is being decided right now by whoever chooses to build the boring thing first.