I’ve argued before that you should give your AI agents a name. I stand by it. Naming works because it forces clarity: once an agent is “Alex” rather than “the tool”, you have to decide its job, what it can access, what good output looks like, and how you’ll know when its performance slips. As change management, it’s the best advice I’ve got.
But I’ve been watching where that advice leads, and there’s a line it must never cross.
The same psychology that makes naming work — thinking of the agent as something like a team member — quietly invites something dangerous. People start to believe the agent can carry accountability. You hear it in the language: the agent decided, the agent got it wrong. Ownership of outcomes drains away, one blamed output at a time.
The concept that crystallised this for me is personhood: the status of being an accountable actor, someone who can be held responsible for what they decide and what follows. Personhood sits with people. It cannot sit with an agent. An agent has no legal standing, cannot own a risk, has no career to lose and no capacity to wear consequences. Underneath the conversational polish, today’s agents are models predicting the next word, paired with tools that let them act. They feel like they’re thinking, so they feel like people. They aren’t.
And the world is starting to blur this line for us. Argentina’s President Javier Milei has put legislation to his Congress creating “non-human corporations”, entities operated by AI agents where human shareholders are optional. Wherever that experiment goes, critical infrastructure must not follow. When the output is wrong, and probabilistic systems will be wrong, a person must be accountable.
So hold both truths. Agents get names, but they go into workflows, not org charts. They have functions, not roles. Owners, not managers. The question is never “why did the agent do that?” It’s “why did the agent I own produce that outcome, and what am I going to do about it?”
Name your agents. Just never promote them.