Who manages the AI?

You have quietly added a team member that does a real share of the work, never sleeps, and answers to no one in particular. Nobody put it on the org chart. That is going to matter.

Here is something I noticed that I cannot stop seeing now.

Every person on your team has a manager, a defined role, and someone accountable for their output. Except the most productive contributor you added this year. The AI writes a real share of the code, drafts the docs, scaffolds the tests, and shapes decisions across the whole team. And it sits on no org chart, reports to no one, and belongs to everyone, which means it belongs to nobody.

We added a worker and changed nothing about how the team is structured around it. That is the gap nobody is designing for, and it is about to start costing teams in ways they will struggle to trace.

Let me lay the question out honestly, because I do not think I have the answer yet. I am increasingly sure most teams have not even asked it.

Start with the simplest version. When AI-generated code fails in production, who owns it?

"The AI wrote it" is not an answer anyone will accept in an incident review, and it should not be. But notice how slippery the real answer has become. The engineer who prompted it and skimmed the output? The reviewer who approved a change they did not fully read because the tests passed? The lead who set the expectation that everyone ship faster? In a fully human workflow, accountability was clear because authorship was clear. AI blurred authorship, and accountability has quietly blurred along with it. Fuzzy accountability is not a small thing. It is the exact condition under which quality rots slowly and nobody can say when it started.

Now the harder one.

For decades, the work AI now does well was the work we handed to junior developers. The straightforward feature. The boilerplate. The first draft someone senior would refine. That work was never really about the output. It was the apprenticeship. It was how a junior built the judgment that turned them into a senior over five or ten years.

AI does that work now, faster and cheaper. So the rational short-term move is to lean on AI and hire fewer juniors. And if enough teams make that rational short-term move, here is the question that should worry every technical leader: where do your seniors come from in ten years? You cannot generate a senior engineer. You can only grow one, slowly, through exactly the work you are now handing to a machine. We may be quietly optimizing away the training ground for the only people who can actually supervise the machine later.

And then there is the part hiding in plain sight.

Someone is already managing the AI. It just is not in anyone's job description. Your senior engineers are spending real hours reviewing, correcting, and redirecting AI output. That is management work. It has the same shape as managing a fast, tireless junior with no judgment of its own, and it is landing on people who were hired to build, not to manage, and whose titles and calendars reflect none of it. (I wrote about the exhaustion of this a few weeks ago. This is the structural version of the same problem.)

So we have a worker with no manager of record, being informally managed by your most expensive people, in time nobody planned for, producing output nobody is clearly accountable for. Write that sentence out about a human hire and you would call it a mess.

I want to be straight with you here. I do not have the clean answer, and nobody I trust does yet.

The teams I see handling it least badly are doing unglamorous, deliberate things. They name an owner for AI-generated work on the critical paths, so accountability has somewhere to land. They are explicit that reviewing AI is real work, and they budget for it instead of pretending it is free. And the sharper ones are protecting junior growth on purpose, deciding which work juniors should still do by hand precisely because AI could do it, because that work was never about the output in the first place.

But these are patches, not a design. The real thing, an org structure built from scratch for teams that are part human and part machine, with roles and ownership and growth paths that actually account for the machine, does not exist yet in any mature form. We are all retrofitting.

The point of this edition is not to hand you a solution. It is to make the question impossible to keep ignoring, because the cost of ignoring it stays invisible right up until it is not. The accountability gap surfaces in an incident. The junior gap surfaces in five years, when your senior bench is thin and you cannot buy your way out of it. The management load surfaces as your best engineers quietly burning out doing a job nobody ever named.

You have a new team member. It is doing a lot of your work. So the question, and I mean it as a real one I am still sitting with myself: who manages the AI on your team, and what is it costing you that nobody has written down yet?

Reply and tell me. I read every one.