The CEOs who are pulling ahead with AI are not asking how much it saves them. They are asking what they can now do for their customers that they couldn’t before. The companies that build adaptive capability fastest will compound. The ones that don’t will be left behind, regardless of how much AI they roll out.
Most companies are still answering the wrong question.
Two people, same company
One person at that company has been quietly using ChatGPT for nine weeks. She has found three things that genuinely change how she works and one that doesn’t. She has told nobody. She is using a personal account because the company hasn’t sanctioned a tool yet, and she has been pasting in things she probably shouldn’t - client briefs, draft proposals, parts of meetings. The productivity gain is enormous, and she doesn’t want to give it up. She knows this isn’t quite right. She also knows the moment she surfaces it, the conversation changes - about her, about governance, about what she should be doing instead.
Another person in the same company has been there for twelve years. He is good at what he does. He can feel the ground moving, and he doesn’t know where to put his foot. He hears the AI message and nods at the right times. Nothing in him has moved.
They look like opposites. They are not. Both want safety and growth, and both are managing the tension by hiding. She hides because what she’s doing is technically against the rules, and she doesn’t want to lose the productivity. He hides because admitting he doesn’t know what to do makes him feel obsolete.
The company has not made it safe to engage with AI honestly. The woman breaking the rules needs the rules to catch up with what she’s already doing. The frozen man needs an honest conversation about where he goes next. Some people need a hand to get started. Others need real guardrails because they’ve run ahead of what the company can check. Each needs something different. The one-size message is why most of this work fails.
Most AI rollouts are answering the wrong question
Most AI rollouts are productivity programmes. Buy the licences, put governance around them, and run the training. Automate some tasks. Hours saved.
That gets you somewhere, but not far enough. Productivity is the easy answer. The harder one is what AI lets you do for the customer that wasn’t possible before. Efficiency is step one. The opportunities AI will create are bigger, and we cannot yet see most of them - our thinking is still anchored to what today’s tools can do.
I have argued in earlier essays that PE often makes a category error - treating complex problems as if they’re merely complicated, reaching for a standard process when the real issue is messier. AI has made that mistake more expensive. But the deeper issue is the question CEOs are asking. How much can we save? is an important question. What can we now do for our customers that we couldn’t before? is the game-changer question.
Stop doing what AI does well
AI is very good at the routine cognitive work. The synthesising, the formatting, the summarising, the analysing. It does that work in seconds.
Here is the uncomfortable truth most CEOs have not yet said out loud: if AI can do something well, your people probably shouldn’t be doing it anymore. One software CEO I work with recently told his engineers, in effect: no more coding, you need to redefine what you’re here to do. The work your people have been doing is better done by AI. Decide what they’ll be valuable for next, and point that work at the customer.
Take Duolingo. They used AI to launch personalised tutoring, role-plays and conversation practice within the app - services the company genuinely couldn’t have provided two years ago. That isn’t a productivity story. The customer is now getting something genuinely new. That’s the question to ask in your business.
What’s left when AI takes the routine work is the part that bookends it. The work now divides into three:
The human decides what’s worth doing - direction, what to prioritise, what matters
AI does the work in the middle — the synthesising, drafting, and analysing
The human reads what comes back - what it means, what’s missing, what to do next
Human at the start. AI in the middle. Human at the end.
That’s where judgement now lives. Not in producing the work, but in framing it before and reading it after. The capacity to push back on AI output. The taste to know what actually matters. The willingness to redefine the problem when the answer comes back wrong.
Your company structure is part of the problem
The company’s structure is the same story. Marketing, sales, product, operations - each running in its own lane. The questions that now matter most cut across all of them. No single function owns them, and that is where most companies are weakest.
You cannot ask people to behave in new ways inside a structure that keeps rewarding the old ones.
You cannot ask your already overwhelmed people to change
You cannot ask overwhelmed people to change.
There is one move that has to come before any of the others. You have to create the capacity for the work.
Capacity has to be made, not assumed. The people most likely to engage are already doing the most. That means killing meetings that don’t earn their place. Cutting reporting cycles that consume more time than they generate. Saying no to projects that would have been yes-by-default a year ago.
But capacity alone won’t do it. You cannot give your people adaptive capability either. They have to build it themselves. What you can do is create the routines that make it possible - regular conversations, protected time, the weekly fifteen minutes that turn a one-off experiment into a habit. The work below is about those routines.
How do you build adaptive capability
Adaptive capability is not a programme you launch. It is built one conversation, one experiment, one redefined role at a time. The companies that build it fastest will compound. Here’s what building it actually looks like.
Aim at a direction, not a destination. You will not know where AI takes your business in three years. Nobody will. What you can know - by paying attention - is the direction. More of what she is quietly doing. Less of what is freezing him. Amplify what’s alive, dampen what’s stuck, keep moving. (I’ve written more on this.)
Ask the customer-value question, not the productivity question. For every function AI touches, the question isn’t how much time we can save. It’s what can we now do for the customer that we couldn’t before? That gives your team something interesting to chase, not just hours saved to count.
Find the experimenters and protect them. In every scaling company, there are people quietly using AI in ways nobody has sanctioned. They aren’t hiding because they want their work to stay private - they’re hiding because what they’re doing is technically against the rules, and they don’t want to lose access. Ask one direct question in a small room: What are you already doing with AI that you haven’t told me about? Then make a promise you have to keep: I’m not going to come down on you. Show me what’s working, and we’ll build the governance around it.
Make the redefinition demand. Pick one team. Tell them honestly: AI can now do most of what we used to ask of you. Come back in 30 days with proposals for the more valuable work. Back the best ones - and remember that testing is a replacement for guessing, not a replacement for judgement.
Concentrate, don’t sprinkle. Find the leverage points where small effort moves the system. Then load resources onto them once you do. Most companies hedge — they spread effort thinly across 30 initiatives because 30 feels safer than 3. It isn’t. When you find the experiment that’s working, the team that’s redefining well, the conversation that’s unfreezing someone - back them properly. Stop most of the rest. The companies that pull ahead aren’t the ones running the most experiments. They’re the ones learning fastest from the fewest. Concentration is uncomfortable. It is also what makes that learning compound.
Give the board confidence about the process, not the result. In complex territory, promising a specific outcome is either guesswork or bullshit. What you can promise is the process. Here are the five experiments we’re running. Here is the hypothesis behind each. Here is how we’ll know what’s working. Here is when you’ll see evidence. The line I encourage CEOs to use, more or less verbatim: “I can’t give you confidence about the result. I can give you confidence about the process, and some confidence that the market is telling us this is where it’s going.” Memorise that sentence.
Tell your people the truth
You cannot ask people to walk towards work that may shrink their current role and pretend that isn’t part of the deal. They know.
The truthful version is harder, and the only one that has a chance. Your people have heard transformation language before. Most of those efforts didn’t deliver. Say so. The short term will be messy. AI will change jobs. Some work will disappear. Nobody can map the medium term from here.
What you can offer is not job security. It is the chance to build judgement, adaptability, and the ability to keep finding what your customer would pay for next — capabilities that travel.
The message, plainly: stay in the game. Keep moving towards the capabilities that are becoming more valuable, not less.
What a CEO is now hired for
I assess CEOs for a living. For most of my career, the quality that best predicted whether a leader would compound value was execution discipline — taking a plan and driving it through an organisation that didn’t naturally want to move. That still matters. It is no longer enough.
What matters more now is harder to assess and harder to develop. Can they hold genuine standards without pretending to have certainty they don’t have? Can they create the conditions where people say what’s actually true, not what’s safe to say? Can they help people keep learning when their sense of competence is what’s under threat?
Those are not questions most CEOs have been asked before. They are the questions AI is now forcing. The lower-bar cognitive work is gone. The standards on what’s left have to be higher, and the honesty has to be sharper. A CEO who can only lead when the path is clear will struggle over the next five years, regardless of how good their AI strategy looks on paper.
And you are no longer above any of it. You are in it with your team - figuring out, alongside them, what’s worth doing now.
The companies that build adaptive capability fastest will compound, commanding a premium at exit. The ones that don’t will be left behind, regardless of how much AI they roll out. Everything else is just motion.