What You Can’t Rent

Most first AI roles are four jobs wearing one title.

Read that last column again. Mostly yes.

A capable internal owner can hold all four. That isn’t the problem. The problem is the last five words of the Build row: one initiative at a time. This tool. That workflow. The automation somebody needs by Friday. Each one can be designed well and run well.

Nobody owns the space between them.

That’s where the cost lives. Two teams maintain the same information and don’t know about each other. Three departments pay for the same capability under different product names. Last quarter’s automation quietly contradicts this quarter’s policy. A workflow is technically successful and creates more work somewhere nobody measured.

None of those problems appears inside a single initiative. All of them appear across initiatives. So the missing function isn’t initiative design. It’s designing the whole — deciding how the pieces fit together, where they conflict, which capabilities should be shared, and what shouldn’t exist at all. It isn’t a fifth box in the job description. It’s the work of making the four boxes cohere.

I went looking for what the market actually does when it creates this seat. I took every US company under 500 people that posted an AI leadership role in the last ninety days and asked two plain questions of the hiring data.

Had they done this before? Nine in ten had not posted an AI leadership role in the preceding eighteen months. For almost all of them, this is the first time.

Were they building a team around it? Seven in ten posted no other AI role in the same window — no AI engineer, no data scientist, no analyst, no architect. One seat.

Put those together and you get the actual shape of the thing. The person you hire walks into a company that has never had this role, so there is nobody inside who can tell them what the job is. And they are not the head of a function. They are the function.

That matters more than the wording of any job description. Four jobs and no team doesn’t divide the work — it queues it. Platform questions, adoption requests, governance calls, and the build somebody needs by Friday all arrive on one calendar, and they arrive in the order other people need them, which is never the order that makes the program coherent.

One more thing worth knowing about the company you’re benchmarking against. Four out of five organizations creating this role have more than 500 people, and nearly a third have more than ten thousand.

The salary comparisons you’re reading are set by that group. If you’re under 500 and you’ve opened this req, you are early — which is a good position, and it means there is no playbook waiting for you inside the building.

If you’ve made the hire, the candidate probably knew more about the subject than anyone on your side of the table. That isn’t a hiring failure — it’s the reason the role exists.

But it means you evaluated what you could judge: how they’d land with your people, whether they’d survive your operating cadence, whether you’d want them in a hard meeting. Those were the right things to examine. They just don’t tell you whether anyone will protect the work between the initiatives.

If you’re still mid-search, this is the cheapest moment you will ever have to fix it. Deciding what the seat is for costs you a paragraph before the offer. It costs a renegotiation after.

And whoever takes the job inherits an operating load that starts at full volume, because people are already doing god-knows-what. Somebody has a personal account they expense. Somebody built an automation three teams now depend on and no one has reviewed. A department is moving company data through a workflow leadership has never seen.

Getting your arms around that is not a distraction from the owner’s job. It is the job. It’s relationship work, political work, and constant work. It runs on standing, continuity, and being findable when someone has a question they’d rather not put in writing. An outsider cannot own it for you. Your internal owner’s hands are full. Correctly so.

Which is why designing the whole goes unprotected — not because the person you appointed is incapable, but because the urgent work of operating the program keeps eating the work of designing it.

The task with no deadline is the task that loses.

The distinction isn’t between an internal operator and an outside genius. It’s between two kinds of work that run on different conditions.

An outside architect should not decide what business you’re in or what your priorities are. The architect’s job is to make the choices and their consequences visible. The internal owner — and above them, the executive team — decides.

Which is why a firm can’t simply replace the internal owner. You can rent design capability. You cannot rent internal accountability.

And you may not need to rent the design capability either. If someone inside already has the range, the experience, and the protected time to do the whole-system work, give them the mandate and make the work explicit. The important thing isn’t that the capability comes from outside. It’s that it exists, and that someone’s week is actually reserved for it.

Find your row: what you have, what may be missing, and what you need next.

If you can already name your row, you have what you came for. Reply to the email that brought you this — or write me at [email protected] — with the row, and I’ll tell you what I’d look at first.

If you’ve already hired, the recommendation to also bring someone in is the most self-serving argument in this brief. So don’t take it on my word. Take it from the work.

Look at the last thirty days. If your internal owner spent the month answering requests, reviewing tools, running sessions, resolving risk questions, and negotiating priorities — the operating function is real and it’s working.

Now look at the space between the initiatives. Who traced a workflow end to end, across departments? Who tested whether a proposed automation should exist at all? Who went looking for duplicated capabilities, contradictory decisions, or a system you’ll regret building? Who named something that should be consolidated, retired, or stopped?

If you can’t name the work product, the function doesn’t exist.

The best internal owner is often not a new hire.

If the job runs on relationships, horse-trading, knowing whose support is needed and where the bodies are buried, the person best positioned for it may already be inside. A COO. An operations lead. A chief of staff. A functional leader with enough authority to set priorities and say no.

They aren’t missing standing. They may be missing technical capability, protected capacity, and a model for how the program should work.

Capability can be built in months. Standing takes years.

If you haven’t started a search, that often makes the cheapest correct answer an internal owner who already knows the company, supported for a while by someone who has designed one of these before. No search. No six-month ramp. No permanent outside dependency.

If the search is already running, the same fact points somewhere else. Whoever you hire will need that standing on day one and won’t have earned it yet — and on the evidence above, they’ll be building it alone. So name the person inside who will lend it: the sponsor who opens doors, settles the priority fights, and absorbs the political cost of the first hard call. That person is part of the mandate. Leaving them unnamed is one of the quieter ways a good hire fails.