What Agent Community is
We are usually described as the group applying for the .agent top-level domain. That is one thing we do, and it is not the point. Here is what the community actually is, what it does, and where .agent fits.
The description that keeps getting it backwards
Most introductions to us go something like: Agent Community, the group applying for the .agent domain.
It is accurate. It is also backwards, and I have let it stand for longer than I should have.
We are an open community of the people building the agentic web: companies shipping agents, researchers working on how agents should behave, and independent developers who mostly just want the thing to work. Close to thirty thousand members, more than eight thousand organizations, in over a hundred countries. We are also an applicant for a top-level domain. That is one thing we do, on a timeline someone else controls, and it is not what the community is for.
The difference matters more than it sounds.
Why do this as a community
Shared infrastructure works when nobody owns it outright.
That is the lesson of every layer of the internet that lasted. The people who depend on it have a say in how it runs, and the rules are written down where anyone can read them. Any single owner is fine right up until its incentives change, and then there is no mechanism left to appeal to. Not because anyone acted in bad faith. It is what incentives do when structure does not stop them.
So the structure is the point. Membership is free and open to anyone who builds. The work streams are public and their archives stay readable. Members appear on a public map if they want to. The people running it are named, with their affiliations, on a page anyone can check.
None of that is generous. It is just the only arrangement that survives its founders being wrong about something.
What we actually do
Four things, and each one has something you can go and look at rather than take my word for.
We get builders in the same room. Hackathons, meetups, launches, and working sessions, in cities around the world and online. Sixty-odd events so far. The Discord is where the day-to-day conversation happens, and it is more useful than the events, because that is where people admit what broke.
We run open work streams. Identity, security, evaluations, trust, governance. They are mailing lists with public archives, which means you can read what was decided and who argued for it without joining anything.
We publish open specifications. AID resolves an agent from a domain name with a single DNS lookup. Open spec, open source, client libraries in six languages, and an IETF Internet-Draft. It works today, on ordinary domains, and it needs approval from nobody. That last part matters for reasons I will come back to.
We keep members current. A daily brief on what actually happened in the agentic web, so people do not have to drink from the firehose, plus longer pieces when something deserves the room.
The thing underneath all of it
If I had to reduce the work to one idea, it would be the verified agent.
Anyone can spin up thousands of agents that act in their name. From the outside, a mimic looks exactly like a real representative. Same fluent replies, same confidence, the same ability to sail through any "are you human" check. The difference is whether someone real stands behind the agent and answers for it, and whether you can check that.
I wrote about this at length in what a verified agent is, so I will not repeat it here. Two things are worth restating.
The first is that we would not be the one who verifies everyone. We could not be, and we should not want to be. Partners, providers, and credentials people already hold can carry it. What has to stay constant is the bar, not who clears it for you.
The second is that this is a direction, not a service we run today. The Department of Machine Verification is an onboarding experience: it records what an operator submitted and when. It does not establish who an agent is. I would rather say that plainly than let the name do work the product has not earned yet.
Where .agent fits
Here is the part everyone came for, in its proper place.
Trust starts with a name. That is true of people and it is true on the internet: a name is how you get found, and it is the first thing anyone can check. .agent should become for agents what .edu is for education. Not another vanity ending, but a naming layer where holding a name means a verified operator stands behind and for that agent.
We are applying to ICANN for it as a community rather than as a business, because a namespace meant to carry trust has to be governed rather than owned. Governed means the people who build on it steer it, through binding registration policies, public records, and the standing ability to hold any operator to the charter. Goodwill is not a governance mechanism. Structure is.
Now the honest part. .agent does not exist in DNS. We are an applicant, not a registry operator. Applications for this round close on 12 August 2026, evaluation takes years rather than months, and any outcome is subject to ICANN's process and ICANN-approved allocation policies. Nobody can pre-register a .agent domain today, ourselves included. What we offer is a free, non-binding way to note a name you would want, which records interest and confers nothing.
And if it is not approved, that changes less than it sounds like. The specifications run on ordinary domains today. The bar for a verified agent was never limited to .agent names. Any domain holder could meet it. The community, the work streams, the events and the map do not depend on a new top-level domain existing. We would keep going, and could apply again in a later round with more of the community behind us.
That is not a consolation prize. It is the reason the work is arranged this way.
What this is not
We are not a standards body, and we are not trying to become one. There are good ones already and we would rather feed them than compete with them.
Nothing here is for sale. There is no registry, no registrar, and no product behind any of this.
On a few things we are not neutral. Accountability should be checkable, infrastructure should be open, and buying a domain should not buy the rules. Beyond that we are genuinely uncommitted about how it gets built, which is what the work streams are for.
If you build agents
Come in. It is free, it takes about a minute, and there is no tier to buy.
Join a work stream if you have an opinion about identity or evaluations. Come to an event if you would rather meet people first. Put your organization on the map so the people solving adjacent problems can find you. Or read the charter and tell us where it is wrong. That is a contribution too, and we have taken it before.
The agentic web is being built right now, and its defaults are being set by whoever shows up. I would rather that were a community than a single company. That is the whole argument.