Your AI problem is not in the technology
Companies are buying AI the way factories once bought electric motors: installed in a way of working built for something else. The return will come from the leaders who change the way of working.
Let me start with something I really believe: artificial intelligence is the most transformative technology most of us will see in our lifetimes. The first time I got scared was when I read an idea that I haven’t had up to that point, it was Ray Kurzweil’s “The Singularity Is Near” in 2005, and its implications for people, and the world at large. Now, as the “singularity is nearer”, as Kurzweil puts it himself (he dates it to 2029), these thoughts keep me thinking at night more often than I would like to admit.
So when I say that most companies are adopting AI in a way that is unlikely to pay off, I am not playing down the technology. I am saying that the problem sits not where most of them are looking.
The saviour on the slide
Jim Collins, in what I think is his most underrated book, How the Mighty Fall, describes a stage he calls “Grasping for Salvation”. The company senses that something is wrong, and leadership reaches for a saviour: an acquisition, a new strategy, a new technology. The saviour itself is rarely the problem, and it often has real potential. What goes wrong is that the organisation adopts it as a story it can tell the board rather than as a commitment that changes how it works.
A lot of what I see happening with AI follows that pattern, if in a milder form. There is a task force. There is a Chief AI Officer, or at least someone who has been handed that title on top of their existing job. Licences have been rolled out, pilots are running, and somewhere there is a slide deck (quite possibly written by AI) with an ambition that would make a human consultant blush, if consultants could blush, that is. Underneath, the way decisions are taken and work is checked carries on much as it did before.
We have seen this before
None of this is new. The most instructive precedent is not the internet but electricity. When the economic historian Paul David studied how American factories adopted electric power, he found that the productivity gains arrived decades after the technology was available and reliable.1 The delay was not technical. Factories had been built around one steam engine driving a central shaft, and for years many owners simply swapped the steam engine for a large electric motor and left the rest of the factory as it was. The gains came only once managers began to redesign the factory itself, with a motor on every machine and the work laid out around the flow of production instead of around the shaft.
Computers went through the same cycle. In 1987, Robert Solow remarked that you could see the computer age everywhere except in the productivity statistics.2 It took years, and a great deal of reorganisation, before the numbers moved.
Most companies are now doing with AI what those factory owners did with the motor: installing it in a layout that was designed for something else.
Where it gets stuck
The current data on AI adoption looks remarkably similar. Atlassian’s State of Teams research for 2026 found that 85 percent of knowledge workers use AI, but only 29 percent have embedded it in how they actually work day to day.3 Atlassian, which has spent two years changing how roughly twelve thousand of its own people work with AI, sums up its experience candidly: “Handing someone a license changes what’s on their screen.”4 It does not change how people decide, coordinate or act on what the tool produces.
What people are willing to say about it is another matter. In an experiment Atlassian ran this spring, knowledge workers judged identical work very differently depending on whether its author admitted to using AI: those who did were rated ten times lazier and were markedly less likely to be recommended for a high-visibility project.5 In companies where AI use is openly celebrated, the penalty almost disappeared. An organisation cannot learn from use its people have good reason to keep quiet about.
Interestingly, the executives who signed off the investment seem to know where the problem is. In The Conference Board’s C-Suite Outlook 2026, among their leading AI priorities, CEOs named building AI expertise in the workforce and improving the culture to adopt it.6 Neither is a technology problem.
So the individuals are using the tools, and the people at the top describe the constraint as a human one. In between sits the layer where adoption stalls: the leaders who run the actual business, and whose teams would have to work differently for any of this to show up in the numbers.
Same same (tools), but different (returns)
This is also why I no longer fully believe an argument I have made myself more than once: that the efficiency gains from AI are strategically worthless because everyone gets them. The argument goes like this. Your competitors buy the very same tools from the very same vendors, so across an industry the net competitive effect is roughly zero.
That is true for access. It is not true for what companies make of it. Every factory could buy an electric motor, but only some of them redesigned how they worked around it, and those are the ones that took the gain.
With AI, I expect the same pattern, only compressed into a few years, because this time the capital arrived first. The money has been spent, and boards have started to ask what they got for it. I would bet that the difference between the companies that can answer that question and those that can’t has very little to do with which model they licensed, and a lot to do with whether the people running the business changed how it runs.
The core is the lever
Here, AI adoption differs from most of the corporate innovation I have worked on over the years. When a company tries to build a new business, its core is usually the obstacle: it pulls budget, attention and people back towards what already works. With AI adoption, it is the other way round. For now, most of the value lies in making the existing business work better, and the existing business already has what that takes: the processes, the customers, the data and, above all, the people who know how the work is really done.
What it often lacks is someone in each part of the business who takes responsibility for redesigning that part. Instead, the responsibility has typically been handed to IT, to a transformation office or to the new Chief AI Officer. They can provide the tools and the guardrails, and they should. But they cannot change how a sales team qualifies its leads or how finance closes the month, because they don’t run those teams.
What it asks of leaders
So what does it actually take? Less than a transformation programme, but more than a training course. As far as I can see, it comes down to leaders doing a few unglamorous things in the part of the business they are responsible for.
The first is deciding what gets redesigned rather than merely sped up. Writing the same report faster is an efficiency gain. Asking whether the report is still needed, now that the analysis behind it takes minutes, is usually where the real return lies.
The second is deciding what good looks like. Which AI outputs are trusted enough to act on, who checks them, and what happens when they are wrong? Researchers at Stanford and BetterUp have coined a term for what happens when nobody decides: “workslop”, work that looks polished and complete but isn’t, and leaves the next person to clean it up.7
The third is making use visible, starting with their own. A leader who uses AI openly, and talks about where it has let them down, does more for adoption than any policy document. Atlassian found that workers who had seen their own manager use AI were four times more likely to keep experimenting with it.8
And the last is owning the result, rather than waiting for someone else to deliver it.
None of this is what these leaders were hired or promoted for. Their role has changed underneath them. In effect, they are going through a leadership transition, only without the promotion: nobody announces it, nobody calls it a transition, and so nobody offers the kind of support we would take for granted if the same person had just been given a new job. (I wrote about the more personal side of this, what it means to lead a change whose destination nobody knows, in The AI questions have changed.)
Not a saviour, but not the problem either
Technology rarely fails to pay off because it doesn’t work. It fails to pay off because the organisation around it keeps working the way it always has, and that only changes when the people leading it change how they lead.
That is the work we do at LUMINUM Partners: with the handful of leaders in an organisation who have to make AI work in the part of the business they run.
