Why .agent has to be a community bid

Every time I go deeper into ICANN's rules, I end up at the same conclusion. There's exactly one shape this project can take without becoming the thing it's meant to replace.

Let me back up.

When you apply for a new top-level domain, you're in a process run by ICANN, the body that coordinates the internet's addressing. Most applicants are companies. They want to operate a TLD because it's a business. Fine. The rules were written for that, mostly.

But tucked inside those rules is something different. A category called community application. If your string is contested (you're not the only one who wants .agent), and you've applied as a community, you can trigger a formal process called Community Priority Evaluation. CPE decides, on the merits, whether your community's claim to the string holds up.

That process is what makes a .agent bid worth running. Without it, the winner is whoever pays the most at auction. With it, the winner can be the people whose work the string actually describes.

This is a bylaws question

ICANN's Bylaws bind the organization to act in the public interest within its technical mission. That's Section 1.1. The first promise the document makes.

Community Priority Evaluation is one of the few concrete places that promise shows up as a rule. It's ICANN saying: a TLD isn't just a thing to sell. It's a naming commitment. When a string describes a real community, that community gets priority, if they can prove it.

That "if they can prove it" is the work.

The scorecard

CPE scores an application against four criteria, with a point budget on each:

• Community Establishment — is the community real, pre-existing, defined? Organization, size, longevity.
• Nexus between the string and the community — does .agent identify the community, or is it a stretch?
• Registration Policies — are there clear rules for who can register, for what, and how they're enforced?
• Community Endorsement — do the organizations that are this community formally support the bid? Who opposes it?

The 2012 round was scored out of 16 points. Applications needed at least 14 to prevail. That bar was hard. Out of roughly 84 community applications, about 38 invoked CPE. Only 5 cleared the threshold.

That's not a discouraging statistic. It's an honest one. CPE isn't a participation trophy. It's a test of whether a community is what it says it is.

What 2012 taught us

Five community applications prevailed in CPE, and all five operate today under Specification 12 community registry agreements with real eligibility restrictions:

• .osaka — 15/16, Interlink Co. Ltd. (Japan)
• .radio — 14/16, European Broadcasting Union
• .hotel — 15/16, HOTEL Top-Level-Domain S.a.r.l.
• .eco — 14/16, Big Room Inc.
• .spa — 14/16, Asia Spa and Wellness Promotion Council

Each had a defined community, a clear nexus between the string and that community, enforceable registration policies, and credible endorsements. Each still does. .hotel's win came with years of challenges: multiple Independent Review Processes and a portal data-breach controversy. It's worth an asterisk if you're studying the record. The other four are clean.

Then there are the losses. .gay was applied for as a community TLD by dotgay LLC. CPE scored them 10/16, twice. The string still became a TLD. Top Level Design won contention privately and runs it today as a standard commercial domain. .music followed the same path: DotMusic lost CPE at 10/16 in early 2016, won the string back at private auction in 2019, and launched in 2022 under a standard base registry agreement. Not a Spec 12 community contract.

That's the pattern when you lose CPE. The string still gets delegated. Just not to a community, and not under community rules.

.bank shows a third path. The American Bankers Association and BITS applied as a community, and the only competing applicant withdrew before CPE was even triggered. .bank operates under a Specification 12 community registry agreement: verified regulated banks only, enforceable eligibility, backed by contract. It's the clearest case of community-by-contract when contention resolves before the evaluation.

So the stakes for .agent are concrete. Win CPE, and we join the small set operating under a Spec 12 agreement with enforceable community rules. Lose it, and the string likely still gets delegated, to whoever outbids everyone else, under whatever policies they set.

The 2012 CPE evaluations were run by the Economist Intelligence Unit. FTI Consulting reviewed the process and published its findings in December 2017. Between the per-application scoring notes and the review itself, the failure patterns become legible. Community definitions that felt real but didn't match the scorecard. Endorsements that weren't from the right organizations. Registration policies that read as aspirational rather than operational.

Every failure point is actionable. Document the community precisely. Match the string to a specific, bounded population. Write policies that would survive a skeptical panel reading them cold. Get endorsements from the organizations whose absence would be conspicuous.

We know what to do. It's just a long list.

What SubPro is changing

The Subsequent Procedures Policy Development Process, known as SubPro, was adopted by the ICANN board in March 2023. It's the rewrite of the rules for the next round of new TLD applications. Implementation is happening in public now, through an Implementation Review Team that started publishing working drafts in December 2024.

We don't know every detail yet. We know the direction. CPE is being refined, not removed. ICANN hasn't yet announced who will run evaluations for the 2026 round, or whether the process will look like 2012's. We'll know more as implementation continues.

For the first application round since 2012, there's an active process shaping how community applications will be judged. The runway is long. The work is front-loaded. You don't assemble a community bid the week before you submit. You do it over years.

The shape of the work

Here's what the scorecard means in practice for .agent, right now.

We have to be a community before we can claim to represent one. Roster, membership structure, governance, longevity — not aspirational, documentable. The string .agent has to read as for this community more than anything else it could be. That's a framing argument we make and rewrite until it's unambiguous. Every rule for who can register a .agent, under what conditions, with what enforcement, needs to be writable on a page and defensible to a skeptical panel. And the organizations that would be expected to support a community-governed agent namespace need to publicly support it. The potential opponents need to be engaged.

Everything we do from here is scorecard prep.

This is a long one

I'm writing this in late February 2025. The application window for the next round is still more than a year out. Evaluations don't happen until after that. If we win, delegation is later still.

I used to find the timeline depressing. Now I find it honest. The rules reward durable work. The process is hard on shortcuts. It's exactly the kind of process you want deciding whether something as foundational as .agent goes to a community or a corporation.

So: we build. In the open, with the scorecard as the plan. If you want to help, there's room.

— Balazs