Essay · June 2026

My Agents Have Names

Naming an agent is the cheapest governance move I know.

I work with four AI agents, and every one of them has a name. Finn handles my work, Pete handles everything personal, Ophelia is my editor, and Sabine runs research. My team once threatened to name our report-writing agent after my mother, who edited with a relentless red pen, though I think they were honoring what her training did to their writing as much as to mine.

The names sound like a gimmick. They are the cheapest piece of governance I have. Giving an agent a name forces a set of decisions you would otherwise leave vague.

For me, the names changed how I work with them. The moment they had names, I stopped treating them as a search box and started treating them as colleagues, and you manage a colleague differently than you query a tool. I have learned where each one is strong, where it is weak, and what it tends to forget. I would not hand Sabine's research to Ophelia to finish, any more than I would hand a research brief to my editor, and when I feel two of them blurring together, that blur is a problem I can go fix.

The names also discipline the work itself. Each agent has a defined patch. Finn stays out of Pete's territory, and Sabine's research does not quietly turn into Ophelia's draft. When I wanted Pete to be more useful, I asked Finn how to move what I had built in one agent's memory across to another, so the context would carry. That kind of move only works once you have decided, deliberately, what each agent is for.

Leave the agents unnamed and you get the opposite: one undifferentiated assistant, every output looking like every other, with no sense of where any of it came from. You cannot tell the research from the draft, the first pass from the verified version, the thing you checked from the thing you only skimmed. The boundaries blur because you never drew them, and you lose track of which output earned your trust and which one is a guess dressed up as an answer.

This is why I treat naming as governance rather than a quirk. An agent has no judgment of its own. Whatever you leave vague, it fills in for you, at speed, every time. The name is where you make the vague explicit: what this one is, what it does, what it does not do, what it remembers and what it loses between sessions. You build that understanding on purpose, the way you would brief a new team member before handing them anything that mattered, because it will not build itself.

A small discipline goes with the names. For each agent, write a one-paragraph remit: what it is for, where its edges are, and what it may not do without you. Hand that to the people who work alongside it, so they know whose output they are looking at. Ten minutes of work, and it does what a name does for any colleague: it tells everyone who they are dealing with.

I started naming my agents because it made them easier to work with, and I kept doing it because it made them easier to govern. Those turned out to be the same thing. An agent nobody has named is an agent nobody is really governing.

So, what are your agents called? If the honest answer is that you just use the AI, that is your starting point.