Leading AI: A motor where the steam engine was, and how task replacement is the least interesting thing about AI.

What's much more interesting for leaders is doing what we could never do before.

On rationing, golf clubs, bedroom pop stars, and the case for leaders with vision over technical depth.

After the panel at Durham University Business School this week, somewhere between the wine and the goodbyes, I found myself telling a fellow guest one of my favourite stories.

It concerns what happened when factories first got electricity, which was, for nearly thirty years, almost nothing.

Mill owners did the obvious thing: they ripped out the colossal steam engine at the heart of the building and installed a colossal electric motor in exactly the same spot - same central shaft along the ceiling, same leather belts dropping to every machine, same layout, same rhythms, minus the coal.

And productivity barely moved for a generation; economists still write papers about those missing decades.

The payoff arrived only when a new wave of engineers noticed: electricity has no centre. Every machine could have its own small motor, so the machines no longer needed to huddle around the shaft. Factories spread out, went single-storey, filled with daylight, and rearranged themselves around the flow of the work itself. Then output leapt.

I keep telling this story because I feel it describes our moment right now in applying AI with uncanny precision.

Most organisations are currently bolting AI into the space where the steam engine used to be: same workflows, same reports, same meetings, minus some typing. That, in a sentence, is task replacement. The disruption it brings is real, particularly in front-line and rules-heavy work, and I don't minimise it.

But it is the smallest and least interesting room in a very large house, and the panel's brief, "Beyond the Hype: How AI will change your leadership journey", is really an invitation to walk through the rest of the house. So do come with me...

Cooking from ration books in an Ottolenghi world

Most of our working habits were designed under rationing. For the whole of our professional history, being informed has been expensive: research cost money, analysis cost time, expertise was scarce. So we built sensible habits around the scarcity, the way wartime cooks built recipes around whatever the shortage left behind. Bread pudding exists because stale ends could not go to waste, and because ingredients were scarce. There is nothing wrong with bread pudding; the trouble starts when you are still planning every meal around stale bread long after the shop shelves have filled up. The pantry is full now. Ottolenghi has arrived with his three kinds of herb and his pomegranate molasses. And a surprising number of us are still, faithfully, cooking bread pudding day in day out.

A client of ours has a stage in their business - as many of yours probably do too - where they conduct some kind of research. The convention in this particular case was to study three comparison companies. Why three? There was never any magic in it. Three was simply the sweet spot of the affordable: enough for a fair comparison, few enough that the study arrived before the decision did, at a price that did not sink the project. Three was a ration, a cost-benefit analysis balanced, and it became best practice.

So the real question, then, is not really whether AI can compare three companies faster. (Prompted correctly, with the right context, yes of course it can - embarrassingly quickly and with far greater rigour.) No, the real question is: why are we stopping at three?

Give that team a steady stream of research agents working through the night, every night, and the ceiling simply lifts. Compare twenty thousand companies, refreshed weekly, daily, hourly if it matters. Then hand the whole living map to the person with the judgement, who slices it by geography, by size, by strategic fit, by whatever the decision in front of them requires, limited only by their ability to think of a question.

In that process, we have gone from a single faded photograph of three companies to livestream footage of an entire market.

And somewhere in that move, the word "informed" changed its definition.

My sense is that this change in definition is worth us all turning in on ourselves, no matter our roles.

Every job carries a duty to be informed before acting; yours does, mine does. So what does "informed" mean to you, in practice, today? Is it reading a few newsletters, the odd paper or publication here and there, scanning headlines daily? Or maybe it's a lot more than that, or less.

Whatever your honest answer, there is a decent chance it was set years ago, by the price of knowledge at the time. But the war ended - the rations no longer apply. Information is accessible, cheap, ubiquitous; it's time to redefine what informed means.

The eight people they were about to let go

Now let us bring this idea of redefining what's possible down to a real business decision that every leader is facing right now, whether they necessarily know it or not.

Picture the team of eight I know whose job it is to write reports for investors. The reporting swallows half their working week - about four full-time people's worth of hours. The engine-room move is obvious: have AI write the reports much as they are written today, using a well-configured Claude project area set up in ten minutes' flat, and take the team from eight to four. The saving is real, the money is in the bank, and for some businesses perhaps that will be the right call. Labour market disruption because of AI is real.

But let us go back to our rationing question for a moment, and a second future for this team of eight opens up:

The costs have not moved in this case; but the quality has risen out of all recognition.

It's leaders' call to make as to which of these futures happen over coming months and years. Which company wins in the market, the one that cut its costs or the one that transformed its client experience, depends entirely on the market, the moment, and what leaders are solving for. There's no single way to embed AI in an organisation; there are many, each pointing at a different future for the individual, team, and business.

Weighing those futures, choosing one on purpose, and then driving the transition with conviction is now the central job of leadership, because the returns at stake outsize anything else on the agenda.

And if that last sentence comes as news, then the first task is simpler still: get informed about what's possible with AI right now, fast!

The washing machine question

One audience question from the evening stayed with me after Nikolaos Mylonopoulos and I answered it in our own ways: are we not becoming too reliant on all this? If AI does the thinking, do our minds not soften? It usually arrives wearing the words "cognitive decline", and it's a common concern, so here's how I think about this:

There are two ways for us to use AI, one being in my view the wrong way, and one, very clearly, the right way. My daughter age 12 puts into her ChatGPT either:

I, as a professional, could put into Claude either:

One of these creates cognitive decline, one strengthens the cognitive process. How are you using AI?

Now the question arrives, does using these tools make us lazy?

I put my clothes in the washing machine every day, and I must say that I don't feel lazy for not carrying them down to the lake with a washboard. I don't know how to wash clothes by hand.

We lose something in every technological shift; that has been true since we stopped making fire by hand. And skills change. We no longer navigate by the stars, and we no longer teach our children to, because it stopped being the one of the highest priority skills their lives depend on. What makes this shift feel different and more unsettling this time is purely its speed: a skill is essential, and then it is gone, within a career or a few months, rather than across a century.

Which points to the skill that actually matters now, the one underneath all the others: the ability to learn, unlearn and relearn, at pace and at scale, using every tool available effectively to do it. As leaders we have to build that capacity in our people while performing it ourselves, live, in public.

It is a tall order.

My fellow panellist Nikolaos Mylonopoulos added the calming other half of the answer: we already lean on technology for nearly everything, and if the digital world went down tomorrow, the physical one would follow within hours. Dependence is not new, and the answer to it has never been abstinence. The answer is contingency: knowing exactly what you would do when the machine stops, and how to get things up and running again quickly.

Wanted: vision and nerve

In 1962, Kennedy stood up at Rice University and said, "We choose to go to the Moon." Much of the technology required did not yet exist; the vision came first, and the engineering assembled itself behind the commitment. I think about that speech often at the moment, because AI compresses that sequence to a degree no previous technology has managed.

The distance between declaring a destination and reaching it has never been shorter, which means the binding constraint on most organisations is no longer engineering capacity. It is the quality of the destinations their leaders can imagine.

So the scarce resource is no longer technical depth; a thousand vendors will explain the models to you before lunch.

The scarce resource is vision, and its inseparable companion, nerve: leaders who can walk their own corridors and tell which walls are load-bearing and which are simply habit, who hear "we study three comparison companies" and ask why three, and who can hold steady through the uncomfortable stretch where the returns are clearly large but their shape is still emerging. The value of this technology is discovered in the doing, and waiting for certainty is usually the most expensive option on the table. MIT reviewed more than three hundred AI deployments last year and found ninety-five per cent had produced no measurable return; the gap to the other five per cent was rarely the model, and almost always whether anyone had built the human capability, and the guided, guardrailed room to experiment, that turns possibility into practice.

Who teaches the drivers?

On stage on Tuesday evening I said that owning an F1 car doesn't make us an F1 driver, and Zoe Maylam kindly quoted it back to me the next morning.

We hold car manufacturers to exacting safety standards, quite rightly, and the same conversation is under way now with OpenAI, Anthropic and the rest, as it should be.

But nobody expects the manufacturer to teach the nation to drive, to issue the licences, or to set everyone's sat nav for them.

They build the machine to standard. Where you take it, and how well you drive it, has always been on us.

Somewhere in the noise, mayhem, and dare I say hype of the past few AI-dominated years, those roles have been conflated. And the conflation is dangerous, because it lets leaders conclude that readiness is somebody else's department; the vendor's, perhaps, or IT's.

But getting people ready to use AI capably and responsibly is the job of every leader, every single one. It is the job of us as parents, as governments, leaders of any kind.

Support in that job has been scarce, and it's been a hard role for us to fulfil. One caution as you go looking for support, borrowed from the rationing shelf: the generic training course is the bread pudding of learning, a recipe invented for scarcity - one teacher, one lesson, served identically to everyone regardless of role, level or how they learn - still on the menu long after the shortage ended.

Capability builds person by person, tuned to each individual and the work actually in front of them; that conviction runs so deep with us, the evidence so clear and so compelling, that we built the entire Atheni.ai platform around it. It only teaches the user exactly what they need to know, at the moment they need to know it, inside their workflow, the way that works best for them.

Barriers, interrupted

One more ration is being lifted, which I find perhaps the most exciting of all: permission.

Think about what it took, until about five minutes ago, to turn an idea into something you could put in front of people. Maybe, like my daughter last week, you decide there should be a cool study app available that gamifies studying and makes it sociable with your friends. You think it might have legs. Until now, you needed developers, or an agency, or ten thousand pounds you did not necessarily have to develop this into an MVP. And with no product or traction yet, you needed access: the warm introduction, the right dinner, the golf club membership. Capital and connections have been, for as long as any of us can remember, the two gates on innovation.

Let's cast our minds back to the 2000s, when the wisdom of crowds was the thrilling new idea. I was building mallowstreet at that time. We marvelled that thousands of Amazon reviewers could steer you to your next book more reliably than one experienced librarian; that a star could rise on YouTube on the love of a crowd rather than a record label's certification. And we were right to marvel; so many barriers lowered.

But crowds only work when there is a crowd, and for founding companies we've had a small one; the only people able to act on an idea involving technology, which is by now pretty well every business idea, were the narrow subset who could raise capital and hire a technical team. It is why the founders we call improbable, the Bransons who left school at sixteen, are famous partly for being so rare.

Economists have even measured what this filtering costs us. The research on "lost Einsteins" found that children from wealthy families are many times more likely to become inventors than equally gifted children from poorer ones; the talent was always there, and the gate was not open.

Let's imagine an economy developing the crème de la crème of ten thousand ideas a day, rather than the best of the few dozen that made it past those two gates.

That would not be just a marginal improvement in fairness, or a small uptick in innovation. That would be floodgates opening. An entirely different growth rate.

Both gates are now swinging open with AI. A founder with a laptop can build a working prototype over a weekend for roughly the price of lunch, research her market to a depth a consultancy would once have billed six figures for, and walk into the meeting with the thing already built and the evidence behind it. She can even have real users on it, collecting the feedback that goes straight into her pitch. She no longer needs anyone's permission to find out whether her idea works. And by the time she has traction and revenue, it's a much more attractive investment.

Music went through this exact door twenty years ago. When making a record required a studio, the labels decided who made music. Then the studio collapsed into a laptop, and the gate simply stopped mattering: Arctic Monkeys built their audience on MySpace before any label had signed off, a Canadian teenager called Justin Bieber was found on YouTube, and Billie Eilish recorded a Grammy-sweeping debut album in her brother's bedroom. Whole genres were invented by people the old system would have stifled.

We have arrived at the bedroom-producer moment for business ideas, and I personally find it thrilling. The talent historically filtered out first - women, people from lower socio-economic backgrounds or minority groups, anyone never handed the warm introduction - is precisely the talent who can now play the game. When the entry fee collapses, innovation stops being a members' club and starts being a field. I for one can't wait to see what grows.

Can someone turn the light on?

Electrification took a generation to pay off because rearranging a factory takes imagination and nerve, and both were scarcer than motors. We have been handed the same choice at a far faster clock speed, with one glorious difference: this time the motors are cheap enough for everyone. You can rewire your own factory today.

Imagine you, electrified.

So, the closing question from the panel this week: optimistic or cautious?

Me, I am somewhere north of optimistic, though not unconditionally so. My optimism rests wholly on my faith in people to step up to the collective responsibility we hold, as parents, as educators, as policymakers, and ultimately as leaders in our own spaces: to have the vision and nerve to notice the ration book in our hands, put it down, and cook something new; to point all this possibility at problems that simply could not be solved before. And to use AI well and teach those we're responsible for to do the same.

There was a fitting symmetry to my week this week. As I travelled home from Northumberland, my co-founder Louise Ballard (Moody) was in Wales filming with international media, capturing the work one of our clients, Grŵp Colegau NPTC Group of Colleges, is doing in one of the most deprived areas of Western Europe: using AI to improve students' learning experience and outcomes, to lift the employability of their graduates, and to ease workload and protect wellbeing for their staff. We have only just begun.

The best uses of any new literacy are invented by the people who acquire it. When reading spread, nobody predicted the novel or the Substack. I for one cannot wait to see what gets written now.

We all have an F1 car each now, whether we have learned to drive it or not. Where we go in it - well, that is up to us.

Have a great rest of the week.

Mackenzie

With thanks to my fellow panellists and hosts: Amir Michael, Claire Ketley, Henry Malton, Jarek Rosinski, Katherine Kirby, Marc Montanari, Nikolaos Mylonopoulos Kieran Jude Fernandes Sue Boyd