I don’t think many people realise it yet, but we’ve just lived through one of the fastest reversals in management thinking I’ve ever seen. It’s only obvious when you look back because it didn’t happen in one dramatic moment. It happened almost imperceptibly until one day I realised I was reading reports from completely different organisations that all seemed to be arriving at remarkably similar conclusions from completely different directions.
For the last two years the conversation had been dominated by the technology. Better models, better prompting, bigger context windows, copilots and then agents. Every few weeks there seemed to be another capability that shifted what looked possible, so it wasn’t irrational to believe transformation would come from the technology itself. The technology really was improving at an extraordinary rate. Looking back, I don’t think that assumption was stupid. I just think it turned out to be incomplete.
Around that time I wrote UNHYPED, largely because I wanted to put a stake in the ground. I wasn’t trying to predict the future. I wanted a permanent record of what I believed while the conversation was still looking somewhere else. I wanted something I couldn’t rewrite later if the market changed its mind. Judging by the meagre sales I’d assume hardly anyone actually read it, although over the last few weeks I’ve caught myself smiling because so much of what I’m now reading feels oddly familiar. Workflows. Operating models. Decision flows. Organisational readiness. Absorption. Those ideas have been running through my writing for the last year. They weren’t where the conversation was then. They increasingly are now.
Over the last few weeks I’ve read reports from McKinsey, IBM, Stanford, Microsoft, banks and private equity firms. None of them are writing the same report. None of them have the same evidence. None of them have the same commercial incentives. Yet underneath all of that they’re converging on a conclusion that felt surprisingly lonely to argue not so very long ago.
The difference isn't the AI itself. It's the organisation around it.
That loneliness is probably why I still remember something somebody said to me recently. “You were super early to this.” They were right. My first thought wasn’t satisfaction. It was, “Yeah... and it was super painful.”
Being early usually is because the easier story almost always wins first. The easier story was that increasingly capable AI would transform organisations. The harder story was that organisations would eventually have to transform themselves before increasingly capable AI could deliver the outcomes everyone was expecting. One story asked organisations to buy better technology. The other asked them to redesign how work and value moved through the business. It isn’t difficult to see which one people preferred.
A few weeks ago somebody asked me whether organisations were finally ready to take on the real work. I found myself saying that I thought some of them probably were, but not because they’d suddenly become more visionary. I think they’ve simply accumulated enough failure that the easier story no longer explains what they’re seeing.
Enough pilots failed to scale. Enough promised ROI never materialised. Enough demonstrations looked incredible until they collided with an ordinary Tuesday morning inside a real business. Enough money has now been spent for boards to stop asking whether the next model will finally solve the problem and start asking whether they’ve been looking in the wrong place all along.
Reality has a habit of winning those arguments eventually.
That’s the reversal I think we’ve just lived through. Not a reversal in the technology, because the technology has continued to improve at a remarkable pace, but a reversal in where we think the solution lives.
For a long time disappointing outcomes were blamed on the technology itself. The models weren’t capable enough. The prompting wasn’t sophisticated enough. The agents weren’t mature enough. Every new release promised to remove another constraint. Gradually those explanations stopped matching what organisations were actually experiencing. The explanation moved. It moved from the model to the organisation.
Buying a better model wasn’t untangling twenty years of accumulated workarounds. It wasn’t clarifying who owned decisions. It wasn’t removing duplicated approvals. It wasn’t redesigning fragmented workflows. If anything it was exposing those problems faster because it has no instinct for navigating ambiguity in the way people have learned to do over years of working around broken systems.
Last September I wrote in UNHYPED that AI doesn’t hide organisational dysfunction, it renders it in higher resolution. I still think that’s true, but I don’t think that’s the interesting part any more. The interesting part is that organisations from completely different starting points are now arriving at the same conclusion independently. They didn’t get there because somebody persuaded them with a better argument. They got there because reality eventually became too expensive to ignore.
I think that’s the real story of the last two years.
The technology didn’t change direction.
The diagnosis did.
That is a much more significant shift than another frontier model overtaking another benchmark because it changes where organisations go looking for answers. If success and failure are primarily explained through the organisation rather than the technology, leadership, workflow design, governance and decision-making stop being implementation details. They become the explanation.
Looking back, I don't think the industry misunderstood the technology nearly as much as it misunderstood where the transformation was going to come from. For a while the assumption was that better AI would compensate for weak organisations. It turned out that weak organisations place a ceiling on what even excellent AI can achieve. The models kept improving. The explanation changed.
That’s why I don’t think the next phase of AI will be defined by another model release.
It will be defined by organisations finally accepting that intelligence was never the hardest part.
Absorbing it was. Redesigning work was. Even if models stopped getting better tomorrow, we’d still have years of work ahead of us.
If the first phase of AI was about making machines more capable, the next phase is about making organisations capable of absorbing them. That’s the argument I wanted to put a stake in the ground for when I wrote UNHYPED. At the time it felt like a lonely place to stand. Looking around today, I’m not so sure it is any more.
Stuart Winter-Tear
Independent AI Advisor | Author of UNHYPED
I agree. I've spent the last 15 years with welding robots, and dropping a robot in the middle of a manufacturing process is no different than stacking AI on top of legacies applications.
Yes. I'm working on a book about this at the moment. I've been tracking both reversals and successes. The difference between the two cases lies in organizational processes, governance, and change management. So many of the reversals were preventable, even obvious, before the FOMO launch. The technology is what it is. Businesses are not being considerate enough about the right use case for AI. And what does good look like? I agree with Lui Sieh below. Many of the same digital transformation issues are recurring with a new technology.