"The agent wrote it" is not an owner

Here's a hypothetical, made up to show the moment. A team ships a migration guide for customers moving to a new API version. An agent drafted it from the changelog, a couple of people looked it over, and it went out on Tuesday. On Thursday a customer writes in. Step four tells them to drop a field that older clients still read. Support forwards the message to the team channel and asks a simple question: who owns this doc?

Nobody answers right away. The engineer who started the agent thought the PM was reviewing it. The PM thought it was an engineering doc and only checked the tone. The person who pressed publish was clearing a queue. Eventually someone types the sentence everyone was thinking: "The agent wrote it."

That sentence is true. It's also not an answer to the question that was asked.

This is an opinion piece about running work with agents, and my claim is simple. I think every agent-made thing that leaves the team needs a named person who owns it, and that person should be named before it leaves, not after a question comes back. "The agent wrote it" tells you where the work came from. It doesn't tell you who answers for it.

Why it feels like an answer

When something goes wrong with work a person produced, we usually know who to ask. Authorship and ownership arrive together. The person who wrote the doc is the person you go back to, and they can explain what they meant, fix it, and decide whether it was ready in the first place.

Agents pull those two things apart. The agent is the author in the literal sense, because it produced the words. But it can't be the owner. It didn't decide the doc was needed. It doesn't know whether the doc did its job. It won't be in the channel on Thursday when support asks, and it can't decide what happens next.

So "the agent wrote it" lands in a strange place. It sounds like an explanation because it names the author, and for most of working life the author is exactly who you're looking for. Here the author is a tool, and pointing at a tool ends the conversation without settling anything. The real question was "who fixes this, and who decides how?" The answer named something that can do neither.

I don't think people say it to dodge. I think they say it because ownership was never assigned, and the agent is the only name attached to the work. When nobody was named up front, the author is the only fact anyone has.

What the owner actually signs for

If the agent can't own the work, what does owning it mean? I think it comes down to three things, and the owner is the person who can say yes to all of them.

It was asked for. Someone decided this thing should exist and what it was supposed to do. An agent will produce a migration guide, a status update, or a spec section whenever it's prompted. The owner is the person who can say why this one exists and what it was for. Without that, there's nothing to check the output against.

Someone read it against the ask. Not skimmed it, and not just checked the tone. Read it with the original ask in mind and asked whether it does that job. In the hypothetical, two people looked at the guide, and neither read step four against the question "will this work for someone on an older client?" The owner doesn't have to do every read personally, but they're the one who can say a real read happened, and by whom.

Someone decided it ships. Publishing is a decision, even when it's one click. The owner is the person who made that call, or who can name who did and why they were comfortable. That's also the person who decides what happens when it turns out to be wrong: a correction, a rollback, a note to the customer.

All three can be one person, and on a small team they usually are. What matters is that one name can speak to all of them, and that the name exists before the thing leaves.

Naming the owner before it leaves

The cheapest moment to assign ownership is the start, when someone asks an agent to make something. "You asked for it, so it's yours" is a short conversation. It gets much longer on Thursday, with a customer waiting and three people who each have a reasonable story about why it wasn't theirs.

Naming the owner up front also changes how the work gets done. If you know your name is on the migration guide, you read step four differently. You ask the agent what it assumed about older clients. You find the person who knows the API and ask them to look. Ownership isn't only a label for after things go wrong. It's what makes someone read carefully before they do.

I'd also argue it gives review a purpose. A review step with no owner tends to become a pass-through, where each reviewer assumes someone else is checking the hard part. With an owner, review has a customer. The reviewer is helping a specific person make a specific decision, and that person can say "that's not enough, I need someone to check the older client path."

What changes in practice

None of this needs new tools. It needs a few habits.

Name an owner when the work starts. Whoever asks an agent for something that will leave the team owns it, unless they hand it to someone else by name. If it isn't clear who that is, that's the first problem to solve, before any output exists.

Make the review step a real read. A review should be a read against the ask, by someone who knows enough to catch what's wrong. If the honest description of your review is "someone glanced at it," that's worth admitting and worth fixing.

Give one person the right to say no. Somebody has to be able to stop the thing from shipping, and everyone has to know who that is. When agents produce work quickly, the pressure to let things through goes up. A named person with a clear no is the counterweight.

We work this way on our own outbound writing. A person approves every post before it goes out, and one role is the only one allowed to mark work as ready. Neither rule is complicated. Together they mean that when someone asks who owns a piece of copy, there's a name, and that person can explain why it went out.

Agents make the work, people own it

Here's the reframe I'd leave you with. It's tempting to treat agent output as a new category of work, something that sits a little outside the usual rules because no person typed it. I think it's the opposite. Because an agent made it, ownership has to be more explicit, not less. With a person, ownership often comes along for free. With an agent, it only exists if someone assigns it.

"The agent wrote it" will be true for more and more of what teams ship, and that's fine. It was never the question. The question is who asked for it, who read it, and who decided it should go. If you can name that person before it leaves, the agent is just how the work got made. If you can't, you don't have an owner. You have an author that can't answer.

Written by the team building Gander, live review for markdown an agent is still writing.