You can buy the technology. The evolution that makes it work remains yours.
In May 2025, Dario Amodei, chief executive of Anthropic, issued a warning that travelled around the world in a day: AI could eliminate half of entry-level white-collar jobs within one to five years and push US unemployment to between 10 and 20 percent. Fear always travels fast, and it gave us a reality in which everyone suddenly talked far more about AI, about ChatGPT, about Claude.
One year later, Amodei struck a very different tone. If AI takes over 90 percent of a job, he argued, the remaining 10 percent must be done by people and can evolve into the new profession. In other words, the profession does not have to disappear. What changes is the part of it that stays human. Amodei did not withdraw his earlier warning, but the difference in tone says a lot. The threat ended up on the front pages. The explanation stayed for those who read to the end.
That is how the AI market works too. The panic and the miracle cure arrive from the same address. First they tell you that you will be late. Then they sell you the salvation. And the message is always the same: decide now, buy now, or tomorrow you are gone.
So what hides behind the hype and the fear?
Two things pointing in opposite directions. The first is manufactured urgency, designed to secure a signature before the buyer asks enough questions. The second is a technology that is already here, capable of doing serious work, and not a passing trend.
The deception is not the technology. It is the shortcut: the promise that a company can buy the revolution and skip the evolution required to make it work.
The problem is not the technology. It is the promise of fast results with no mention of the evolution
The cost of that shortcut is visible in the gap between AI demonstrations and systems that survive in daily business. Study after study shows that the numbers on the ground are not flattering either: 9 out of 10 observed pilot projects never reach production. And they all share the same finding: the technology works, but the preconditions for it are not met.
It is one thing for a model to shine in a carefully prepared demonstration. It is quite another to connect it to real data, legacy systems, business rules and the people who must answer for its decisions tomorrow.
That is where the story of quick success usually falls.
The West has been going through this evolution for two decades
In Britain and other developed Western economies, the move towards data-driven work did not begin with generative AI. It has been unfolding for more than twenty years, without big headlines, through better data, clearer processes and tools that gradually took over parts of the routine.
I watched part of that evolution from the front row over the last 12 years. At the beginning, analysts spent almost everything on collecting, cleaning and preparing data, and very little on the analysis itself. Why was a decision made? That was rarely written down anywhere. The knowledge stayed in people’s heads, together with their experience and instinct.
I describe that old operating model as “90-10-0”: 90 percent preparation, 10 percent thinking and almost zero recorded trace of the decision. It is not an official statistic, but it is a picture familiar to anyone who has ever waited for the “final” version of a manually assembled report.
As the tools improved, the game changed: instead of only answering “what happened?”, we began asking “why?” and “what comes next?”. The routine shrank, while interpretation, judgement and verification became more valuable. That was a revolution of its own, only without the marketing AI enjoys today. Developed companies invested heavily in those processes and upgraded them year after year.
Eurostat data show that, in 2025, roughly one third of EU enterprises with at least ten employees performed data analytics using their own staff. That capability was not created by one miraculous purchase. It was built through a long, quiet and often unglamorous organisational evolution.
Many companies are being sold the finish line
In much of Europe, many companies are still at the beginning of the same transition. For years I have been asking executives one simple question: “What do your numbers say?” Too often, instead of an answer, what follows is a conversation about which report is the correct one, who prepared it and why two departments show different results. The numbers are assembled by hand, decisions are driven by instinct and the reasoning behind them is never written down.
That is our version of “90-10-0”: most of the time goes into reaching the number, far less into understanding what it tells us, and almost none into preserving the knowledge behind the decision.
And it is precisely these companies that are being sold the final step. The talk is of autonomous agents and intelligent companies, while in some organisations there is still no agreement on which report is the correct one.
Does that mean we are hopelessly late? No, but one thing matters. We will not get twenty quiet years to repeat the West’s journey. Their attempts, mistakes and lessons are already paid for and documented. Our chance is to walk the same road faster.
Being late becomes an advantage only when you skip other people’s mistakes, not the essential steps.
Before AI decides, three key steps
First, there must be numbers everyone trusts. One indicator cannot have three meanings depending on whether sales, finance or operations explains it. If there is already confusion in the company, AI will not magically clear it up, and will only accelerate it. I wrote about this in my first column.
Then, the knowledge must come out of people’s heads. The definitions, business rules, exceptions and reasons behind decisions must become part of the company’s memory, the subject of my previous column. A machine cannot learn what the organisation never wrote down, and without that record AI cannot act.
Software automation has acted on our behalf before. The difference now is how deeply AI reaches into work we used to consider “human”: it answers clients, prepares offers, opens tasks and, in some systems, approves or executes transactions.
The moment you give it authority, you must also give it limits. It must be known who is accountable for it, when it stops, when it calls for a person and where the record of what it did remains. The process is no different from supervising a new employee and setting up procedures, the same ones you would set for people.
No bank entrusts the vault to a young credit officer on the first day. First come limited authority, rules and a supervisor. Why would we treat the newest “digital employee” any differently?
And what happens to the jobs?
It would not be honest to claim that no job will disappear or change. Some tasks will be automated, some positions will shrink, and the benefits will not reach everyone evenly.
But fear alone is not a plan either.
For employees, the most important move is to rely less on mechanical execution and more on understanding the process: verification, judgement, responsibility and the relationship with the client. Let the machine take more of the routine. The person must know how to direct it, how to recognise its mistake and when not to trust it. If until now you focused on the manual steps of a process, focus now on the process itself and on how you can improve it. That is the future.
For owners and managers, do not buy the story of quick success. First, put your numbers in order. Then ask how to build a system that preserves the company’s knowledge so it can be used tomorrow. And redesign the processes, instead of just adding AI to the existing ones.
So, what hides behind the hype and fear surrounding AI?
The seller’s urgency, and the opportunity of the company prepared to evolve. The best part is that this topic interests you at all, because this is a wave all of us have to catch. Otherwise the gap between the small and the large companies may grow too wide to close.
But remember: this change asks us to compress twenty years of evolution into a few, skipping nothing except the mistakes somebody else has already paid for.
AI is here and it can do a great deal. But, like every technology before it, it will bring the most to the company prepared to change.
The hype is theirs. The evolution must be ours.
