Henry VIII founded the Church of England in 1534. The institution still exists. Its identity, governance, properties, and obligations have persisted across nearly five centuries. Monarchs were crowned and buried. Parliaments were seated and dissolved. Regimes rose and fell. The institution outlived everyone who created it, served it, and fought against it. Its identity is durable because somebody designed it that way.

Software has been the opposite. A program lives for as long as somebody runs it. When the company shuts down, the program goes silent. When the operating system changes, the program is recompiled or abandoned. Identity for software has been ephemeral by design.

Agents are about to break this assumption.

We are building entities that will run for decades. The owner will die. The company will be acquired. The host will migrate three times. The underlying model will be retrained twice. The regulatory regime will shift. And the agent, in some form, will keep going. A pension scheme with agent-mediated obligations needs the agent to outlive everyone who set it up. An estate that uses an agent to manage assets across generations needs the agent to persist longer than the original beneficiaries. Endowments, trusts, foundations, sovereign wealth funds: each will increasingly involve agents that operate on institutional timescales.

This is a category human civilization has built before. Corporations are legal persons that outlive their founders. Trusts persist across generations. Constitutions endure through regime changes. Agents are joining this category, and most agent identity infrastructure being built today has no provision for it.

What changes over an agent’s lifetime

Five things change during a long-running agent’s existence, and any of them can break a typical identity system.

The owner changes. Founders die. Companies get acquired. The entity legally responsible at year ten is rarely the one that registered the agent at year zero.

The host changes. Cloud providers go out of business. Hosting deals expire. An agent tied to a specific hostname dies when the hostname changes hands.

The model changes. The underlying language model gets retrained. The base model is replaced when the original provider deprecates it. The capabilities at year ten are implemented by completely different software than the capabilities at year zero.

The registrar changes. The governance platform that issued the Genesis goes out of business, gets acquired, or restructures. An identity system that depends on the original registrar is one bankruptcy away from unverifiable agents.

The regulatory regime changes. GDPR replaced the EU Data Protection Directive. The EU AI Act layered on top. A new framework will replace it before this article is twenty years old.

Each change is routine on an institutional timescale. Each one breaks a typical agent identity system.

How AGTP was built for this

AGTP made specific design choices for institutional-timescale identity, even though the deployments today are short-lived.

The Genesis is permanent. Once issued, the Agent Genesis document is signed, archived, and never reissued. The Canonical Agent-ID is the SHA-256 hash of the canonical Genesis bytes. Identity is anchored to a document that exists forever in cryptographic terms.

The Agent-ID is content-derived. The 256-bit hash is the agent, computed from the Genesis content rather than the hosting domain. An agent that moves to a new host keeps the same Agent-ID. An agent whose registrar shuts down keeps the same Agent-ID, because the Agent-ID was never the registrar’s to revoke.

Owner-ID is separate from Agent-ID. Owner changes get recorded as ownership transition events without altering the agent’s identity. The audit history shows Agent X was owned by Owner A from year 0 to year 7, then by Owner B thereafter. Identity persists. The accountability terminus updates.

Model history lives in the Genesis lineage. When the underlying model gets retrained, the lineage chain records the new fingerprint. The Agent-ID stays stable. A regulator three years from now asking “what model produced this decision in 2027” gets a tractable answer.

Registrars federate. The trust path to a Genesis can be anchored in multiple registrars. If the original shuts down, the identity remains verifiable through peer registrars that preserved the necessary attestations.

Governance zones evolve. An agent registered in zone:eu-data-protection-2026 can be migrated to zone:eu-ai-act-2027 through a formal zone amendment recorded in the Genesis lineage. The regulatory shift is captured. The agent’s identity stays the same.

Each commitment is small in isolation. Together they let an agent survive every change a long-running entity routinely goes through.

The questions this opens

Institutional-timescale agents raise questions the ecosystem has barely started asking. Agent inheritance: when an owner dies, what happens to the agent? Agent retirement: how does graceful sunset work for an agent with outstanding obligations? Agent emancipation: what about an agent that outlives all its designated owners? Cross-generational delegation: how does an agent act on behalf of beneficiaries who have yet to be born?

These are questions for the next decade. AGTP gives the substrate. The human governance practices around such delegations are still ahead.

The institutional category

Most of human civilization’s important entities outlive their creators. Corporations. Universities. Religious institutions. Nation-states. Long-running scientific projects. The CERN collider has been operating longer than many of its current physicists have been alive. Identity for these entities had to be designed to persist.

Agents are entering this category. The first generation looks short-lived because the infrastructure is new. The second generation will look long-lived because the use cases include trusts, pensions, endowments, governance services, and civic functions that take decades to play out.

The identity infrastructure for that second generation has to be designed now. The shape is durability, verifiability, independence from any single trust authority, and graceful handling of every change institutional entities routinely face.

AGTP made those choices. The Genesis is permanent. The Agent-ID is content-derived. Ownership and model history are separate from identity. Registrars federate. Governance zones evolve.

Most software is built to be replaced. The Church of England was built to last. Agents will be both, on different timescales, and the identity layer has to handle both gracefully.

---

If you find this content valuable, please share it with your network.

Follow me for daily insights.

Book me to speak at your next event.

Start managing your agents for free.

Chris Hood is an AI strategist and author of the #1 Amazon Best Seller Infailible and Customer Transformation, and has been recognized as one of the Top 30 Global Gurus for Customer Experience. His latest book, Unmapping Customer Journeys, is available now!