Why your most important AI hire is already in the building

76% of CEOs in IBM's 2026 study have, or are hiring, a chief AI officer. 57% of those CAIOs are promoted from inside. Most AI deployments don't fail at the model. They fail because no operator was named to inherit them.

Published: 2026-05-19

Author: Ahmed Heshmat

Category: Field notes

Reading time: 7 min

Key takeaways

• The chief AI officer role barely existed two years ago. 26% of large companies had one in 2024. 76% have one or are hiring this year. That curve does not move without something real behind it.

• Inside those same companies, 86% of employees have the skills to use AI; only 25% use it daily. A 61-point gap. The bottleneck is not literacy. It is ownership.

• In IBM's 2025 survey of 600 chief AI officers, 57% were promoted from inside. The person who closed the gap was already in the building before the title existed.

• For operators evaluating AI: the highest-leverage person on your team is not the one you hire next. It is the one already doing the job who picks up AI faster than her peers. Find her before the model arrives.

The pilot is not the hire

We get called in after pilots die. We have written before about [why most AI pilots die after the demo](/blog/why-most-ai-pilots-die-after-the-demo) — four pre-conditions, all of them violated by a clean pilot deck. One of those pre-conditions matters more than the others: somebody on the client's side has to own the system on the Tuesday morning after launch. By name. With time on their calendar.

In the engagements we have shipped that held, that person was always already inside the company before we showed up. Sometimes she was the operations manager. Sometimes a senior coordinator. Once it was a receptionist who had spent her weekends building Notion templates the rest of the staff used. None of them had "AI" in their title. All of them did the work that the title would eventually formalize.

The pattern is consistent enough now that we have stopped treating it as a coincidence.

The number that keeps coming up

IBM published the survey behind this in early 2026: 2,000 CEOs from publicly traded companies with median annual revenue of about $5.8 billion. Globally, the numbers are almost certainly softer than that — this is not the world, this is the top of the world. The directional truth is harder to argue with.

Two years ago, 26% of those companies had a chief AI officer. This year, 76% have one or are hiring. The role moved from rare to standard in twenty-four months — about an order of magnitude faster than the chief information security officer arc after the internet hit.

Inside those same companies, 86% of employees have the skills to use AI in their daily work, or could pick them up with light training. Only 25% actually do. A 61-point gap.

This is the number we keep coming back to. Not because the precise figures matter — they are self-reported survey data, and the real gap is somewhere in that neighbourhood but not exactly there. What matters is that the gap exists at all, that every CEO running these companies knows it exists, and that nobody has built the bridge.

The bridge is not training. Training has been the default answer for two years and has not closed the gap. The bridge is ownership. Somebody has to be named, by management, as the person responsible for closing it. Without that name on a calendar entry, the 86% does not become the 25%. The hours sit on the table.

The role is being invented in real time

The chief AI officer role does not have a template yet. 40% of CAIOs in IBM's 2025 survey reported to the CEO. 24% reported to the CIO. The rest were scattered across CTOs, COOs, and other senior leaders. No two of those reporting lines mean the same thing. That looks chaotic. It is actually the most useful fact in the report.

A role with no template means the first person who defines what it looks like inside a given organization wins by default. There is no playbook for them to be compared against. If you walk into an operating meeting six months from now with the version of the role that actually works inside your company, you are the playbook.

This is also why the path through the title matters less than the path through the work. 57% of CAIOs at major companies were promoted from inside. They were already operating as the AI person in the building before HR caught up. The title arrived after the work did.

That order — work first, title later — is how this kind of role gets created in every industry. It happened with the CISO after the internet. It is happening with the CAIO now. The companies who get this right notice the person already doing the work and put a budget behind her. The companies who get it wrong hire an outsider with a clean LinkedIn and watch them spend nine months trying to understand the operation.

What this means for operators

Two things follow from this for the kind of small and mid-sized operators we work with, which is mostly property management, brokerages, and trades businesses in Toronto. Neither of them is "you need a chief AI officer." Almost none of you do, and pretending otherwise is theatre.

First, the highest-leverage person on your team is not the one you would name if we asked you who runs things. It is the one in the third row of the org chart who is already using ChatGPT, Claude, or some workflow she rigged in Make on her lunch break. She is the one who will inherit whatever we build. Identify her before we start. Bring her into the audit. She will save you from the deployments that look elegant on paper and collapse on a Wednesday because nobody on the operating team understands them.

Second, the question we ask first when we scope a build is not "what would be most impressive." It is "who, by name, is going to be running this on the Tuesday morning after we hand it off?" That question is the operate phase of [our audit-build-operate model](/blog/audit-build-operate-model), and it is the reason the model is structured the way it is. The named internal owner is not a nice-to-have. The deployment lives or dies on whether that person exists, was identified early, and was built around.

If the answer to "who owns this on Tuesday morning?" is "we'll figure that out later," we tell the client the same thing every time: we should not start the build yet. Not because we are precious about it. Because we have watched the alternative enough times to know how it ends.

What we won't do

We will not deploy a system in an organization where no operator has been named to inherit it.

We will not let an executive sponsor act as the operating owner. Executive attention is too thin to maintain a system through its first six months. The owner has to be inside the operation, with the operation's incentives.

We will not run an audit without sitting with the people doing the work. The internal owner is almost never the person we get introduced to first. We have to find her ourselves. That is part of the work, not a delay against it.

These are not innovations. They are descriptions of what it actually takes to make AI hold inside an operation that was not designed for it. We arrived at them the way we arrived at most of [the discipline we named the company after](/blog/why-we-named-the-company-the-system) — by watching the alternatives fail.

The model is the easy part. The Tuesday morning is the entire job.

The point

Two years ago, the chief AI officer role barely existed. Today, 76% of large companies have one or are hiring. The most overlooked fact in that curve is not how fast it moved. It is that more than half of those seats were filled by people who were already in the building, doing the work, before the title existed.

The same pattern shows up at the operator scale we work with. The AI-native version of your operation is already half-built by somebody on your team who is not waiting for permission. She is the operator the deployment is going to rely on. Find her before the model arrives.

If you are in a regulated sector — healthcare, finance, municipal — the leverage on this is even higher. Domain knowledge plus AI fluency inside a real constraint is the rarest hire in the market right now. You almost certainly already have the first half on your team. The second half is something we can help with.

The model was never the hard part. It never will be.

نظام. One deployment at a time.

---

Sources. IBM Institute for Business Value, 2026 CEO Study (n ≈ 2,000 CEOs of publicly traded companies). IBM Institute for Business Value, 2025 Chief AI Officer Study (n ≈ 600 CAIOs).