Promoted, not replaced

Why the real frontier of AI is people, not models

When I came to San Francisco last April 2026, the shared excitement was all directed at Agentic AI.

Everywhere, on the streets, ads for companies in the AI supply chain. All born with the goal of building the technological stack of the agentic organization, without doubt the new organizational model that companies today should be looking toward, so as not to fall behind.

Recently, the extraordinary improvement of LLM models was making it possible to give those intelligences "arms," so you could imagine automating every kind of knowledge worker task. The idea behind it is in fact very simple: if you equip a super intelligence with the same tools that you, human, use every day, you'll soon find yourself no longer having to carry out tasks. In this context, all you'll need to do is give direction to the machine, and maybe evaluate its work.

Back to us 3 months ago, time runs incredibly fast in the SF bubble, the more forward-looking companies were already rewriting their processes to enable new ways of working.

I remember clearly people telling me they saw the possibility of cutting their team by more than half the people, delegating the repetitive work to the machine. 3 months ago San Francisco saw in the cost-reduction use case the very same use case that now seems to interest our clients in Italy and Europe (keep this point in mind, because we'll come back to it shortly).

Fast forward, 3 months later, the situation on the street is still dominated by the Agentic AI revolution.

This time, though, together with Claudio Bedino, our Head of AI Transformation, we talked with these companies much more, at length and in person.

And the perception is that pretty much all of them do the same thing. Building agents, building data layers, monitoring and observability.

Telling who the winner will be among the endless array of companies being born is incredibly difficult, even though each one tries to find its original edge to enter the market.

This brings us to the first two considerations.

The first won't surprise you. If it's true that to understand the future you have to look at where investments go, then we can say with certainty that AI will really represent an epochal transformation in the way we perceive work. Anyone who still thinks that all in all nothing is changing doesn't realize that entire categories of workers are already now giving way to hyper-specialized agents that do most of that work with fewer errors and at a lower price.

The second is that the presence of so many similar companies on the application layer of AI is the sign of a technological scenario that hasn't defined itself at all yet.

Also, whereas last April Anthropic's supremacy on frontier models was clear, after just a few months the situation is already changing.

The release of Kimi (a Chinese open-weight model) clearly instills fear in Western AI Labs, for its ability to almost match the performance of Fable (Anthropic's spearhead) at a third of the cost.

This tells us that it's not on LLM models that we should focus our attention. By now they're commodities, so much so that as soon as a new one is published, before long the others reach the same performance.

And it's not even on the application layer, as I was saying, that we should focus. Figuring out which agent builder will take the market isn't all that relevant.

What is relevant instead is what sits at the extremes of the supply chain. That is, in the infrastructure, datacenters, chips, cloud, and, lo and behold, in adoption.

The new real frontier of AI is its Adoption

The thing that surprised me most of all is the enormous welcome that Wibo, the company I co-founded, received in conversations with the protagonists of San Francisco technology.

Wibo, for those who don't know, deals with Work Transformation. Essentially we map how people work. We intercept dysfunctional human behaviors, involving the interaction between people (e.g. the manager who struggles to manage their team) and also between people and Artificial Intelligence. We then also intercept the dysfunctional processes, the ones that can be improved by inserting AI Agents, and we work to implement them.

So, once we've identified these dysfunctions, through training and the transformation of processes we make sure the new technology is used well. In a functional way.

That way, essentially, we do change management. Last mile adoption of artificial intelligence, as they called us here in SF.

Adoption, to my surprise, is the strongest concern I heard in San Francisco.

The surprise is tied to the fact that SF is a city of builders, developers. None of them ever worried much about whether their technology got used, and used well. These are people paid several hundred thousand dollars (millions, many times) to, in fact, have fun. They're paid to build things we'd call "nerdy," frontier technology. Their concern certainly doesn't lie in whether ordinary people then use this technology.

And despite this, in every past technological revolution, somehow the technology reached everyone anyway, effortlessly. Think for example of the very fast spread of social media, by word of mouth. Or of the cloud: it was enough to invest in huge armies of account executives to bring the hyperscalers (Google, Microsoft, AWS) into every company.

Yet today, faced with this new technology, Generative AI, adoption particularly worries even the builders.

And even more so the investors. I had the luck of speaking with a few General Partners of mega funds invested in AI. And it's clear that after putting so much money into this technology, the concern immediately goes to AI starting to produce, massively and in the real economy, the promised results.

It's not just about adding, the way it used to happen (a new website on top of the physical presence, an e-commerce on top of traditional distribution, a digital social network on top of personal relationships), here it's about radically transforming the way we perceive work.

And, as a consequence, the very relationship we have with work. Changing ourselves is not easy at all. It requires an enormous effort, questioning the system of habits that brought us this far. Accepting that a machine does a substantial part of our current job much better than we do.

That a developer writes code more slowly than an AI.

That a doctor is more inefficient at making a diagnosis than an AI.

That a lawyer is slower at document research and drafting contracts than an AI.

That a human contact center creates more frustration than one based on AI.

When I feel resistance to these statements, strong ones I realize, I think of that famous video of Steve Ballmer (former CEO of Microsoft) mocking the new iPhone because of the absence of physical keys. A few years later iPhone was everywhere and defined a new standard, proof that every technological leap always follows the same pattern. It stays incomprehensible to most until it becomes inevitable.

I could go on for a long time.

Adoption inevitably faces resistance. This is why the existence of companies like Wibo meets with so much favor from Silicon Valley operators.

It's not just about using the technology, though. It's about using it to have, ultimately, a business return. An impact on the P&L, so to speak. And this is missing, not only because to date the use (and even before that, the deployment) of agents is absolutely behind, but also because once brought into the org chart, these AI employees have to be useful to the business.

Reduce costs or increase revenue?

In this scenario, reducing costs is the first of the use cases that organizations see. When I talk with our clients it's clearly the first thought that jumps to their mind.

And it's also what comes to mind for every single worker, who inevitably ends up seeing themselves suddenly out of a job.

But, I'm sorry to say, stopping at cost cutting is a shortsighted reasoning. In fact reducing costs is not the opportunity offered by this revolution. It's, at most, a side effect.

And it's the conclusion everyone would reach if they reasoned with the head of a history expert instead of the head of someone who does management control.

The history of technology has almost never been a story of cost reduction. Not in the capitalist context we live in, certainly.

It has always been, instead, a story of growth. Every time a technology arrives capable of creating new capabilities, the world doesn't spend less. It spends more, and gets much more.

For example, in 1979 VisiCalc was published, the first electronic spreadsheet. It did in one second what a bookkeeper could do in a whole day, without errors, by the way. It looked like the end of an era. And in fact in the United States the jobs for accounting clerks, the ones who spent their time lining up numbers, fell by about four hundred thousand. But the jobs for accountants grew by about six hundred thousand. Because when doing the math became a commodity, people started asking for other things. Once a problem, a bottleneck, is overcome, humanity moved on to problems of greater scale. So technological progress moves the bar of problems upward, requiring a shift of the human presence onto new fronts.

Another example I like to recall concerns an American truck driver, Malcom McLean, who in 1956 first put goods inside standard metal boxes, containers. Before his innovation, loading a ship cost a few dollars per ton. With the container the cost collapsed to a few cents. And here comes the point. Because that initial cost optimization did not serve to ship the same things spending less. In fact people started shipping much more and global trade exploded. So entire countries, emerging economies, entered the world economy because all of a sudden moving goods cost almost nothing.

What I've described is the effect of technologies that change the course of history, as Generative AI probably will. They open the door to new jobs you couldn't even afford to imagine before.

And you don't have to go back decades. It's happening now with software.

Until yesterday writing software was the bottleneck. It cost time, engineers, months. Today that cost has collapsed. An engineer who uses these tools well produces ten times what they produced before. Writing code is no longer the problem. And with the previous bottleneck gone, the work, instead of ending, simply moves.

So now the bottleneck is data and context. Knowing what you want to build, for whom, inside which process. The human part, as it happens.

In parallel, what you can afford changes too. Before you bought pre-packaged software and adapted the company to it. Now you can build it to measure, tailored to the exact way you work. One piece taken from a tool you like, one piece from another. These are elements that in recent years companies had maybe decided to outsource and that they now bring back inside.

The opportunity is therefore not given by cost reduction, but by all that is left on the table today because the organization doesn't have the arms to gather it.

There's a further example brought to me by one of our Executive Teachers, concerning his experience and confirming what I've said.

It's a family office, over a billion and a half under management, with the inbox full of investment opportunities, but no person with enough time to open all the emails to explore those opportunities. Which inevitably get lost.

It's clear that inserting AI into this process, for example an agent to do email triage and catch the important ones, has nothing to do with reducing costs, but rather with spotting the investment opportunities that would otherwise stay there because there are literally no people who would read them.

To understand the problem of AI technology adoption you have to flip these reasonings onto the companies we too work with every day.

How many are the missed opportunities?

The employee who could grow and doesn't grow. The product that doesn't get launched for lack of capacity. The market that could be segmented in a better way to refine marketing campaigns and that doesn't get segmented simply because there's no time. Those are the opportunities that AI would help capitalize on.

So why do we see almost only the cost side?

Scarcity vs abundance mentality

I believe it's because we look at the world with the eyes of scarcity, instead of those of abundance. In Italy, then, where competition is much less fierce than in the US, companies are less afraid.

And when you're not afraid of losing, you stop playing to win. You play to protect.

The scarcity game is always the same: instead of aiming high and improving, you play defense. Instead of creating, you optimize. It's simply a zero-sum game, and nothing new has ever come out of a zero-sum game.

So the most paradoxical thing happens. A company that sees AI only as cutting ends up with one more line in the P&L, the cost of tokens, that wasn't there before.

It pays the people, it pays the technology, and it doesn't see the revenue line move upward. That's how you explain the great disillusionment in AI adoption, the one now immobilizing a lot of organizations. We aimed this technology toward optimization instead of toward opportunities.

Abundance, far from being a matter of LLM models or agents, is instead a matter of people. Organizations grow if people grow. And people don't grow if they feel replaced by a machine.

This is the real, ultimate obstacle to adoption. More than the technology itself, it's about fear.

If you ask anyone to work side by side with something that does their job better than them, faster and without errors, you'll see that the natural reaction will be defense. Rejection. Scarcity applied to oneself.

The only adoption that works is the one that instead flips this feeling. That doesn't make people feel replaced, but promoted. Lifted from a lower level to a higher one. Freed from the task to return to judgment, to taste, to direction. To the things that stay ours.

This is exactly my work at Wibo. Which is not bringing agents into companies. We're not system integrators and not AI consultants either. Our work is to bring people higher, to elevate the human, to give life to a new form of anthropocentrism.

In which the person, at the center, is enabled into a new condition.

In which the person acquires new human behaviors, more functional ones, that we train.

I like to think of myself as that company that brings people a step higher by teaching human and AI skills to get them there. We map how people work to understand where they get stuck and we transform that friction point into a promotion.

Because in the end AI doesn't serve to make us work less.

Some of the people who work the most in the world are based in San Francisco. And if they had already found the way to remove work for good, they'd already have done it.

AI serves today (maybe in the next essay I'll correct myself) to make us work better.

And a person who works better doesn't feel replaced.

They feel promoted.