Q What does the data tell you about where most organizations actually are with AI right now?

A recent study from Harris Poll came out that should capture every global CEO’s attention. Eighty percent say their role is on the line if AI fails to deliver by the end of 2026, yet forty percent are not present for most of the decisions that will determine whether it does. Having seen this pattern play out across different industries and technology cycles, the result tends to look the same: accountability travels upward faster than decision-making authority does, and the distance between the two is where things go wrong.

What tends to fill that distance is a significant amount of handwaving. Leaders expect AI accountability to flow downward, either through a technology team or center of excellence, and then express frustration when the outcomes do not materialize. The frustration is justified, but it is pointed in the wrong direction. The honest question to ask any CEO in that situation is when was the last time you personally used AI, and what did you use it for? The answer usually tells you everything you need to know about how seriously the organization is actually treating AI, and whether they are prepared to close that gap or simply manage around it.

Q Why does AI keep becoming someone else's problem?

This is very similar to what happened with big data. Companies sent their technology teams off on multi-year journeys to build data lakes and platforms. The cost and complexity went up, and at the end nobody had a clear idea what they were going to do with what had been built and there was no path to value. It is not a sophisticated analogy, but Mike Tyson said it best: everyone has a plan until they get punched in the mouth.”

MIT’s Project NANDA found essentially the same story playing out with AI right now. Despite $30 to $40 billion in enterprise spending in 2025, 95% of generative AI initiatives produced no measurable impact on the bottom line. Klarna replaced roughly 700 service agents with an AI assistant, then rehired people once quality slipped and the promised savings never occurred. The technology functioned as expected, but what failed was the belief that buying a capability is the same as absorbing it into how the organization actually works.

In most organizations, the instinct in to stand up a dedicated group, treat the work as a project with a start date and a finish line, and wait for a handoff that is never going to come. It is like trying to fit a Ferrari engine into a Camry. You can build something technically impressive, but if it cannot be absorbed into how people work and it sits alongside the business rather than inside it, nothing actually changes.

The tools that exist today make building AI capabilities remarkably fast and effective. What does not get easier is changing the workflows around those capabilities, and that work cannot be handed off to a task force or a vendor. It has to be owned by the business at every level, and that ownership has to start with the CEO.

Q How should CEOs think about AI differently than other technology investments?

It is important for CEOs to understand that the outcomes from AI do not come from any single piece of it. The data has to be clean and governed. The workflows have to be meticulously redesigned. People need both the skills and a real reason to change how they work. And the budget has to account for the fact that usage costs scale as adoption grows. Remove any one of those and the whole system cracks before it has a chance to build momentum. And those workstreams need to build on each other. Better data produces better outputs, which draw in more people, who surface the real use cases that make the next round better. Partial commitment breaks that loop early, and organizations rarely see the ROI.

A CEO who treats AI as one more project to staff and finish is only managing only a piece of the problem. The one who positions it as the way the business now operates is managing the full picture, and those organizations tend to pull ahead of the ones stuck in project mode. The project mode companies share a common trait: they measure activity rather than outcomes. They can tell you how many pilots they have run and how many tools they have deployed, but they cannot tell you whether any of it has changed how the business actually performs. That is the difference between managing an AI program and leading a transformation.

Q Why does every function need its own AI answer?

AI is far more a business initiative than a technology one, and treating it as the latter is one of the most common and costly mistakes organizations make. When it lives in the technology function, it gets treated like a technology problem, something to build and hand back when ready. When it lives in the business, it forces a different set of questions. How does this change what we are able to do? What do our people need to know to work well alongside it? How will we know if it is actually moving the needle? Those questions can only be answered at the function level, and the CEO is the only person with the standing to require every function leader to answer them.

Every function is going to have a different answer to what AI means for its work, and that needs to be by design. A group where expert judgment drives most of the value is going to approach this differently than one where the work is more repeatable, and trying to run both at the same pace with the same expectations is how you end up with one function racing ahead while another is still figuring out what the question is. What the CEO can require, across all of them, is that every function leader is genuinely engaged and has a working answer for where they are placing bets and what impact they expect to see.

There is also a timing problem most organizations are not paying enough attention to. AI plans are too often being built around what the models could do twelve months ago, so function leaders plans end up being under-built for what’s truly possible. Add to that the 6 month timeline it takes many corporations to move from idea to pilot, and companies are positioning to be falling behind from day one. The right question to be asking is not what can we do today but what do we want to be able to do in two or three years, and how do we need to reorganize and tap into the latest models to get there.

Q What does governance actually mean for a CEO, and why does it matter now?

Governance tends to get treated as the boring part, and when the conversation around you is about what AI can now do and what is coming next, stopping to talk about guardrails and documentation feels like pulling over on a highway. But if a CEO cannot answer basic questions about what AI is actually doing across the organization, they are accountable for outcomes they have no real way to see or influence. The questions a CEO should be able to answer to demonstrate they are in command of their AI strategy are not complicated. Where are we using this? How are we measuring whether it is working? What do we do when something goes wrong?

The governance problem is also becoming increasingly measurable. Gartner expects more than 40% of agentic AI projects to be shut down by the end of 2027, indicating organizations deploy agents without a clear plan and without the controls to know what happens when one does something nobody expected. A CEO does not need to understand how an agent works technically, but they do need to be able to answer what the organization is letting these systems do on their own, and who is responsible when one of them gets it wrong. The organizations that get clear on that early will actually move faster, because they can safely give their agents more room to operate.

Q How should CEOs think about ROI when so much of AI's impact is hard to measure?

A lot of executives worry about picking the wrong AI vendors, and that makes sense given how many big technology bets have gone sideways before. But that concern tends to focus on the wrong variable. Picking the right vendor matters, but the harder question is whether the organization has a real path to return on what it has already spent. And in my experience, most organizations do not have a clear answer to that, partly because it is hard to measure and partly because nobody wants to admit it.

The return comes from how people and processes actually change around the technology, not from the technology itself. Organizations keep treating the tool as the answer and then are genuinely surprised when the results do not follow. If the work of changing how people do their jobs has not been planned and funded, the investment will not return what was expected regardless of which vendor built the underlying capability. This is the same thing that happened with the data lake. Everyone felt good about building it, and then at the end of it there was nothing to show.

Most organizations are not going to be able to run a clean before-and-after comparison because too much else is changing at the same time. What a CEO can do is require a clear picture of where things stand today, agree on what the organization is trying to change, and track the early signals that show whether behavior is actually shifting. A consistent scorecard over time tells you far more than any single metric, and building that baseline is something most organizations skip because it feels unglamorous compared to the work of building things. But every time a CEO accepts a story about AI impact in place of actual evidence of it, the gap between accountability and results gets a little wider.

Q What does it actually look like when a CEO is genuinely leading AI across their organization?

That survey finding, where 80% of CEOs are accountable for AI outcomes but 40% are absent from the decisions that determine them, does not close just through better frameworks or more layers of process. It closes when the CEO is holding every function leader to the same standard the board is holding them to. Every function leader should be able to walk into a leadership meeting and make the case for where they are placing their AI bets, why those areas, and what progress looks like over the next year. The important distinction is making the case rather than giving a status update, defending a business position they own and are prepared to be held to.

The CEOs who are actually making progress on this are using AI themselves, and using it on real decisions. Bringing a real capital allocation question or using it to pressure-test a recommendation is the only way to develop a real feel for where the technology works well and where it falls short. Without it, a CEO has no real basis to push back on the work their function leaders bring them, and they end up nodding at confident answers they have no way to evaluate.

AI has been around since the 1960s, and it is easy to forget that when it feels like everything is moving at once. We have seen wave after wave of this. Every decade or so a new capability arrives that people insist is different, and the organizations that figure out how to absorb it rather than just install it are the ones that pull ahead. The leadership principles that held through all of those waves still apply. Set a direction. Put real resources behind it. Hold people to it, and model the behavior you want to see. What is different now is that those things have to be applied to AI by the CEO personally, kept up continuously, and not treated as something to delegate and revisit quarterly.

The CEOs that get ahead of this are those who have recognized that leading AI is simply part of the job now.