Give Every AI Agent a Name

The fastest way to lose the thread on AI isn't a rogue model. It's an anonymous one.

Here's a test I've started running with founders. How many AI agents are operating inside your company right now — and who owns each one?

Almost nobody can answer. There's a pause, a rough guess, and then the honest admission that they're not really sure. That pause is the entire governance problem in miniature. You cannot govern what you cannot name.

The reason is structural, and it's worth sitting with. When a person joins your company, they arrive with an identity. A name. A manager. A role with a scope. A record that grows as they work. All of that scaffolding is what makes them accountable — it's how you know who did what, and who to talk to when something needs to change.

An agent arrives with none of it. It shows up as a workflow someone wired together on a Tuesday afternoon because it was useful. No name. No owner. No written record of what it's allowed to touch. And then it just starts acting — drafting, sending, deciding, moving things — inside your operation, anonymously.

Multiply that by a dozen and you see the real risk. It was never a rogue superintelligence plotting in the dark. It's a handful of unremarkable agents, each doing one small thing, none of them owned, quietly compounding into a system nobody can see the shape of and nobody can answer for. Diffuse, blameless, everywhere. That's not a sci-fi threat. It's a Tuesday.

The fix is almost embarrassingly analog: give every agent a name, a human owner, and a record. A name so you can talk about it, find it, and audit it. An owner — a specific person, per the last piece in this series — whose name is attached to what it produces. And a record of what it did and on whose authority it did it.

I'm not being cute with the “name” thing. This is exactly where the serious frameworks are landing. Singapore's agentic-AI governance framework, published this year, requires every agent to carry a verifiable identity and an audit log of who authorized its actions. Large enterprises are standing up “agent registries” for the same reason. The instinct behind all of it is right, even if the machinery is overkill for a company your size.

Because here's the right-sized version, and it fits on one screen. Keep a single registry — a shared sheet, one row per agent. Five columns: the agent's name, what it does, who owns it, what it's allowed to decide on its own, and where its actions are logged. That's the whole thing. And one rule to go with it: if an agent isn't in the registry, it doesn't run. New agent, new row, before it touches anything real.

What surprises people is how much the naming alone changes behavior. The moment an agent has a name and an owner, it stops being “the AI”, that diffuse, blameless force everyone shrugs at, and becomes a specific thing a specific person is accountable for. Accountability is a spotlight. Anonymity is a fog. You'd be amazed how much cleaner people are with a tool the moment their name is next to it.

The record matters most on the worst day. When something goes wrong, a bad number, a customer email that shouldn't have gone out, a decision nobody remembers making, the question is always “which agent did this, and who authorized it?” With a registry and a log, that's a ten-second lookup. Without one, it's a forensic investigation across five tools while the problem keeps running. The trail is cheap to build now and priceless the day you need it, which is precisely why nobody builds it until the day they needed it yesterday.

You would never let an anonymous contractor with no manager and no paper trail make decisions inside your company for months on end. An unnamed, unowned agent is exactly that, you've just been calling it “automation” instead of what it is.

So give it a name. Give it an owner. Give it a record. That's not bureaucracy. It's the difference between having a workforce and having a fog.