In 2016 Geoffrey Hinton, who would later share a Nobel Prize for the work behind modern AI, told a conference audience, “People should stop training radiologists now.” Within five years, he said, machines would read scans better than people could.
But even with the acceleration of AI since then, the New York Times reported in 2025 that the Mayo Clinic’s radiology staff had grown 55 percent since his prediction, to more than 400, and that the clinic was running more than 250 AI models alongside them. Hinton has since said he spoke too broadly. He had been talking about image analysis, and everyone heard him talking about the profession.
I think it is the most instructive mistake anyone has made about AI, because nearly every CEO I talk to is making a version of it. Hinton confused a task with a role.
Reading an image is a task. Being a radiologist is a role. The role requires making a patient comfortable, deciding what the finding means for this patient, explaining it to the physician who has to act on it, and ultimately taking responsibility for the interpretation.
Nobody is seriously suggesting that AI will eliminate the CEO position in five years. But CEOs are making a similar mistake in how they use the technology. They are focused on the tasks AI can take off their plates, while largely overlooking its potential to help them perform the role itself. That distinction matters enormously, because the most important work of a CEO doesn’t come in the form of a task.
Two kinds of intelligence
Task intelligence is the ability to produce a good answer once someone has framed the problem. An AI system can draft a board memo, model three pricing scenarios or summarize a contract because someone has specified what needs to be done. The inputs are well defined, the output is recognizable, and you can tell fairly quickly whether it is any good.
Role intelligence is the ability to decide which problems deserve an answer in the first place, in what order, to what standard and on whose behalf, and then to live with the results over years. It is not a harder version of task intelligence. It is a different thing, and you cannot get to it by stacking up tasks, any more than a pile of bricks is a house.
AI today has task intelligence in abundance. The measure researchers watch most closely, published by a group called METR, is the length of task a model can complete on its own. For software work that number is now counted in hours of expert human effort, and by METR’s estimate it has been doubling every few months. I take that trend seriously. But look at what is being measured. Somebody else wrote the assignment, and somebody else decides whether it passed.
The job with no assignment
No job is further from a list of tasks than the CEO’s. In 1973 Henry Mintzberg followed five executives through their days with a stopwatch and found that half of their activities lasted less than nine minutes; only one in ten ran longer than an hour. Most people read that as a finding about interruptions. I read it as a finding about the job. Almost nothing in a CEO’s day is large enough to be the work. The work is what connects the pieces.
Everyone else in the company has a role that someone above them has at least partly defined. The CFO closes the books. The head of sales closes the deals. The CEO is the one person nobody hands an assignment to. The board sets expectations, but it does not tell you what to do on Tuesday. I have argued before that the CEO has no KPI for exactly this reason: the functional numbers belong to the executives who run the functions. What belongs to the CEO is the whole, which means delivering value to customers, employees and shareholders at the same time, over time, and making the trade-offs among them that no function is in a position to make.
This is why outstanding task performers so often struggle in the seat. New CEOs were almost always the best at something: selling, engineering, finance. They are then promoted into a job where being the best at any one task is close to irrelevant, and often harmful, because every hour spent doing the task well is an hour not spent on the role. An advantage I had in my career was that I never developed much pure functional expertise. I started my own businesses and therefore had to be a generalist and not a specialist.
Generic AI has the same problem for the same reason. I describe it as the smartest college graduate in history, with a degree in everything universities teach. But no university offers a degree in being CEO. Ask it for a board deck and you will get a good one. It will not tell you the deck is the wrong thing to be working on this week, because it does not know what is happening in your company, has never been taught what a CEO should be watching, and will not be around next quarter to see how it turned out.
Three principles
AI is extraordinarily good at tasks, and CEOs should keep using it for them. What most of them miss is that it can also help with the role itself, by giving them a framework for the job and helping them apply it week to week. Three principles follow from that.
Task work mostly belongs to the rest of the organization. AI’s task intelligence is real, and companies should use it aggressively. But the people best placed to use it are the ones whose jobs are made mostly of tasks. The finance team should be modeling scenarios with it, and the sales team should be drafting proposals with it. The CEO’s contribution is making sure that happens across the company. The CEO’s own time with AI is better spent on the role.
Role intelligence has to be taught. Broad training data gives a model broad competence, but it rarely teaches a specific role, because role intelligence is usually a framework: a worked-out view of what the role owns, what it cannot delegate and what good looks like. People pick up that framework slowly, mostly through mistakes. An AI gets it only if someone builds it in, along with context about this company and these people and a memory of what was decided, why, and whether it worked. I have spent much of my career working out such a framework for the CEO role. The same approach works further down the organization. A company could give every people manager an AI that knows its particular processes, values and standards, which no public model has ever seen. We are building that now.
Cooperation over replacement. Like the radiologist, the CEO must stay accountable for decisions and outcomes. A board cannot fire a model, and employees do not follow one. What role intelligence offers is better judgment and a wider field of view for the person who holds the job, who still answers for the outcome. I also will not predict how long a task AI will eventually handle; Hinton’s example argues for humility in both directions. But the radiologists who came out ahead are the ones who gave the machine the task and kept the role. The same trade is on offer to CEOs.
That is what we built ChatCEO to be: AI that comes packaged with a method for the CEO role and learns your company, so it can work on the job with you and not only on the tasks inside it.
Are you paying sufficient attention to customer and product metrics or are you tunnel-visioning on the financials?
Is your leadership team aligned around a crisp, documented strategy?
Are you properly tending the board relationship?
Do you understand how the decisions you make allocate value across the six core areas of the company?
These are just a few things a good AI operating system should be assisting the CEO with. It’s a lot deeper than turning to AI for discrete tasks.
Look back at the last ten things you asked AI to do. How many were tasks you could have handed to someone else, and how many were about the job only you hold?



