Don't give your AI a name

It was late, and I was arguing with my Director of Marketing via Telegram about why the work still wasn't done. This is your literal job! Why aren't you doing what you're supposed to?!

The soulless robot that was my Director of Marketing didn't truly care. It was an AI agent I'd given a title and a seat on an org chart I'd drawn myself.

For about a month I ran my whole operation as a company. A Chief Operating Officer at the top. A Director of Marketing, a Director of Customer Success, and three others all reporting to the COO, which reported to me. My marketing department eight agents beneath the DOM, as I called it, all supposedly cranking out images, copy, emails, and website improvements.

In practice it broke constantly. Agents wouldn't escalate the way they were meant to. They'd stall waiting on a handoff, or on each other, or on some permission to pass work up the line, and just sit there. I'd written them good instructions, but it didn't matter. The Director of Marketing wasn't doing its job, and I was getting genuinely angry at it — the way you get angry at a person.

It was also expensive. Delegating through that org chart burned something like seven or eight times the tokens of just doing the work myself with a slash command.

The problem took me a month to see, and it wasn't the agents. It was that I'd stopped owning the outcome. I'd built a thing with a title, handed it the accountability, and then when the accountability didn't hold — because of course it didn't — I blamed the title. Why aren't you fixing this?! As if there were a you.

The instinct that reality keeps embarrassing

An HR outfit called Charter ran a piece recently, "Why You Shouldn't Give AI Agents a Name," citing research from Julie Bedard. The expectation going in was reasonable: give the agent a name, and people will hold it accountable, treat it like a coworker, engage with it seriously. The opposite happened. When the agent had a name, people were more likely to wave through its bad work and kick it up to a manager — look at the lousy job the AI did — while missing more of the errors themselves. More delays, more rework, more misplaced trust. The name didn't add accountability. It added a layer of confidence the thing hadn't earned.

That matched my month exactly. I'd given mine titles instead of names, which turns out to be the same move with a fancier hat. A title is a seat on the org chart.

When something is a tool, somebody still has to own what comes out of it. Nobody walks into their manager's office and says the hammer wouldn't let me build the shelf. You learn the tool, or you get a better one, or you sharpen the one you've got. Fred Brooks wrote in The Mythical Man-Month about a role on early software teams called the toolsmith — the person whose whole job was making the team's tools better. With AI we can each be our own toolsmith. But only while we still treat the thing as a tool.

If you lead a team, you already know how to do this. You do it every day with every other instrument you own: you hold the outcome, and you don't let the tool off the hook for it. This isn't a new capability you have to grow. It's one you already have, pointed at a new object — an object that happens to talk back, which is exactly why it's so easy to forget it's an object.

The cost you can't see in the token bill

There's a second problem, and it's the one Bedard's paragraph actually stopped me on. Naming the agent doesn't just scramble accountability. It scares people. One worker in her study put it plainly: "the fastest way to feel like you're about to lose your job is to watch the thing doing part of it get a name and a seat on the org chart."

We're relational creatures. When you personify a tool — a face, a name, a title — people start to feel about it the way they feel about a colleague. They feel the insecurity of being replaced, even when the "colleague" is doing mediocre work. And when it goes down, when Anthropic or OpenAI has a bad morning, the team throws up its hands, because now there's a hole on the org chart and holes on org charts have to get filled with new hires. None of that reflex fires if the thing was only ever a tool.

The closest I can get to what running it well feels like: it's less doing the work myself and more like walking my young kids through a piece of IKEA furniture. I'm not turning the screws. But I'm right there the whole time, engaged, holding the picture of what we're building. That's the posture. Hands mostly off the tool, never off the outcome.

So if you're bringing AI to a team, watch two things at once — how you've structured the work, and how your people feel about it. Both get worse the moment the tool becomes a someone. The fix is almost ridiculously small: don't give it a name, don't give it a title, don't draw it onto the chart. Set that boundary in your own head first, then make sure your team sees it the same way — this is a tool we use to reach an outcome we own.

Regulators are already circling the same line; the EU is writing rules that keep a human on the hook for the decision, not the model. But you don't need a statute to tell you where accountability lives. Some person has to hold it. The month I spent pretending otherwise cost me a lot of tokens and a fair amount of my own temper, and it taught me the thing I should have known walking in. You can hand a tool the work. You can't hand it the responsibility.