Perspective

What 25 Years of Digital Transformation Taught Me About the AI Hype

By Sean Doherty · July 14, 2026

In 1994 I watched a room of senior managers get very excited about a website. It had a spinning logo, a guestbook, and a hit counter that we had quietly padded to make the numbers look respectable. Everyone agreed it was the future. Nobody in that room could tell you what the site was for, who it served, or what would happen the day after launch. I remember thinking, even then, that the spinning logo was doing an enormous amount of persuasive work and none of it was honest. That instinct has paid my mortgage ever since.

I have now lived through four of these waves, and I am watching the fifth arrive. The web came first, then mobile, then social, then cloud, and each one landed in more or less the same shape. First a demo that made everyone gasp. Then a stampede of companies buying the thing before they understood the problem it solved. Then a long, unglamorous stretch where a smaller group of people quietly did the boring work and won. The gasp and the win almost never came from the same place, and they almost never came from the same company.

Think about who actually took each wave. The web was not won by the firms with the flashiest Flash intros. It was won by the ones who rebuilt their catalogue, their fulfilment, and their customer service around the fact that a stranger could now reach them at three in the morning. Mobile was not won by whoever shipped the first app. Half the early apps were a shrunken website in a wrapper, and they died. It was won by the people who understood that the phone was a different context, a different posture, a different set of thumbs, and who redesigned the whole encounter accordingly. Social was drowning in brands opening accounts because a consultant told them to have a presence, and the ones who mattered were the handful who treated it as a listening instrument rather than a broadcast tower. Cloud, the least photogenic of the lot, was won almost entirely in the plumbing. No demo, no gasp, just a slow migration that changed a company's cost structure and its speed of experiment. The pattern is embarrassingly consistent once you have seen it a few times. The flashiest demo is a decoy. The value hides in the operating model behind it.

So when people ask me whether AI is different, my answer is yes and no, and the no is the useful part. The technology genuinely is a step change. I am not one of the people standing at the back muttering that it is all autocomplete and hot air. I have built fifteen or so enterprise AI proofs of concept, and a good number of them did things I would have called impossible a decade ago. I am bullish, and I have the receipts. But the way people are buying it, and the reasons they give themselves for buying it, are identical to 1994. The spinning logo is back. It just talks now.

Here is the thesis I have arrived at, and I will defend it against most of the room. The model is almost never the hard part. The hard part is the human system you drop the model into. I have watched brilliant models fail because nobody could agree who owned the output. I have watched mediocre models succeed wildly because someone had done the tedious work of cleaning the data, defining the decision, and teaching forty people to trust the thing and to know when not to. The intelligence in the box gets all the press. The intelligence around the box is what determines whether you have built something durable or an expensive party trick.

This is why the demo is such a reliable liar. A demo is a controlled environment with clean inputs and a friendly presenter steering you away from the edges. Your business is an uncontrolled environment with filthy inputs and nobody steering. The gap between those two things is not a technology gap. It is an organisational one, and no model release will close it for you.

The demo answers the question can it work. The only question worth paying for is will it keep working on a wet Tuesday when the person who understood it has left.

When someone brings me an AI initiative now, I do not start with the model, the benchmark, or the vendor's slide deck. I start by asking a small number of deliberately unromantic questions. What decision or task does this actually change, named specifically enough that we could measure it. Who is accountable for the output when it is wrong, because it will be wrong, and a system with no owner for its mistakes is not a system, it is a liability with good PR. What does the human do differently on Monday morning, and have we asked that human, or are we about to ambush them. And what happens on the bad day, when the input is garbled or the model is confidently mistaken, because the quality of an AI initiative is decided almost entirely at the edges, not in the happy path the demo showed you. Notice that none of those questions are about the model. Three of the four are about people. That ratio is not an accident. It is the whole lesson of twenty five years compressed into a checklist.

I will admit this is not a flattering message to deliver. Everybody wants the wave to be about the technology, because technology you can buy. You cannot buy your way out of an ambiguous decision right or an unclear line of accountability. Those require the unfashionable work of sitting with the people who do the job and understanding it before you automate it. The firms that will win this wave are the ones already doing that work, mostly in private, without a press release. Same as every other wave. The gasp is happening in public. The winning is happening in the plumbing.

Which brings me back to where I started, and to the one discipline that has never once let me down. In 1987 I became Toyota's first Kaizen engineer, and I built their first intranet before most people had a word for what an intranet was. Kaizen is not complicated. Make a small improvement. Prove it works. Then, and only then, scale it. That is the entire philosophy, and it is quietly devastating when you apply it to a hype cycle, because it is immune to gasps. A spinning logo cannot survive the question prove it. Neither can a chatbot that impresses in the demo and falls over on the wet Tuesday. Kaizen forces you to find the smallest real thing, ship it into the actual mess of your actual business, measure whether it moved the number you named, and expand only from proof. It is the natural predator of hype, and it works on every wave because it never cared which wave you were on.

I find it faintly funny that the most powerful technology of my career is best handled with a shop-floor discipline from the 1980s. But the more capable the model, the more expensive it is to scale the wrong thing quickly, and the more valuable it becomes to have someone in the room whose first instinct is to shrink the ambition, prove the increment, and refuse to be dazzled. I have been that person for a quarter of a century. The logos kept spinning. I kept asking them to prove it. So far the arithmetic has held.