Before Agent Identity Had a Name: What Is Actually Datable, and Why We Claim No Prescience
The thing had no name because the pieces were separate
Software has been proving what it is to other software for decades. Certificates issued to servers, accounts that belong to a process rather than a person, and systems that hand a running workload a short-lived credential.
None of that was called agent identity, and the properties that distinguish it are measured elsewhere in this estate rather than re-argued here. The short version: the older field solved lifetime, and it did not solve portability across parties who share no infrastructure.
What is actually datable, and what it shows about us
The standard's own contracts repository reports a creation timestamp in October 2025.
Our surface for this was scaffolded at the end of May 2026. Our software package began in July 2026. The first passport anchor was attempted in mid-July 2026.
So we arrived roughly eight months after the standard's repository existed. That is the unflattering ordering and it is the true one.
The upstream date comes from a public registry API and ours come from our own history, two different systems.
And the first day did not go well
The anchor log's first day carries forty failed registrations, all dated to that one day, with the successes beginning the same day and continuing later.
That story is told with its numbers elsewhere, and it belongs in a history page because the shape of a real beginning is a day of failures followed by a fix, not a launch.
Why this page claims no foresight
A history page written by a company is usually an argument that the company saw something early.
We did not see this early. The dates above rule it out, and any version of this page implying otherwise would be contradicted by our own commit log.
What we can say is narrower and checkable: the identifier work is genuinely older than the agent work. The method this surface uses was being built and submitted for registration before the agent surface existed, and its registry entry was merged in early July 2026.
That is not prescience about agents. It is having built a general thing that turned out to apply, which is a much more ordinary and more defensible claim.
What is real, because absences alone would misdescribe us
A passport anchored in a public registry, readable without an account.
An identifier method registered in the W3C DID Method Registry, with the entry merged and checkable in one request.
And a published package with framework adapters, which is the ordinary work of showing up rather than the extraordinary work of having been first.
What actually changed when the name arrived
A name lets separate people build compatible things. Before it, everybody solving this solved it privately, so nothing composed.
The standard's registries are the visible effect: a place to put an identity that is not any one vendor's database.
And what has not changed: the hard parts are still unsolved. No published deployment for the registry that would hold evidence of a check, no agreed vocabulary for capabilities, and no resolver driver for our own method. A name organised the conversation; it did not finish it.