You can't trust an agent from a company that has no root
An AI agent's identity is only as good as the company root it points back to. Without a corporate root there is no way to validate an agent's authority or measure its output.
Everyone is racing to give AI agents an identity. Most of that work starts in the wrong place.
Attackers today rarely need to steal a key. They use a system that is already allowed to use it, and every signature checks out. A valid signature proves a key was used. It doesn’t prove the company intended it, or consented to it.
An agent’s identity is only as good as the thing it points back to. When an agent shows up at your bank, your supplier or your customer’s systems and says “I act for Acme,” the only question that matters is: says who? If Acme has no root (no digital name of its own, under its own keys, that publishes who may act for it), then there is nothing to check. The agent’s credential points to a vendor’s directory, or a login, or a key someone generated last Tuesday. That isn’t authority. It’s admission.
And without a root you can’t measure output either. A measurement has to be signed by someone accountable. If the company itself can’t sign, every claim about what its agents produced is just a claim.
So the first building block of the AI-first company isn’t an agent platform. It’s a corporate root:
- A digital name the company creates for itself. It isn’t issued by a vendor, and it isn’t rented. It’s held under the company’s own keys, with more than one signer, so no single person or vendor can act as the company.
- A public record of who may act for it. Which people, which agents, for what, and until when.
- Revocation that takes effect everywhere at once. When someone leaves or an agent misbehaves, the company withdraws the authority in one place, and every counterparty can see it.
The root belongs to the board, not to IT. It’s the company’s authority, so it should sit with the directors and follow the legal and accounting structures the company already has.
This isn’t hypothetical. Rootz Corp’s own digital name is a 2-of-3 wallet on Ethereum, registered as a Doing Business As name in Delaware. It has been publishing signed authorisations since 2024: which keys may act for the company, and for what, from crypto wallets to certificates. Every key now has a ledger, and the ledger is what makes it possible to validate a key and to revoke it. Read our authorisations, or ask your AI to read them: corpid.rootz.global/verify/rootz.
As David Birch put it last week, know your business has to come before know your agent.
Every agent, every receipt and every measurement points back to a root. Companies that don’t have one will find their agents can’t be checked, and agents that can’t be checked won’t be let in.
First in a short series on how I see AI working from here. Rootz is building, partnering and investing in this future every day.
The series: how AI works from here
- 1. You can't trust an agent from a company that has no root
- 2. Logs are for IT. Receipts are for AI.
- 3. Running unknown compute is no longer an option
- 4. If your company doesn't speak AI, your customers' agents can't use you
- 5. The asset you're giving away is your context