# Co-intelligence: when AI can do almost everything, the real talent is knowing how to judge

25 October 2026 · Alessandro Caprai, AI educator, trainer and consultant

<https://caprai.dev/en/editorials/post/co-intelligence-when-ai-can-do-almost-everything-the-real-talent-is-knowing-how-to-judge>

---

There was a time when, to learn how to judge others' work, you first had to do it yourself, and judgment was the slow fruit of experience, mistakes, and effort.

Today, a machine does that effort in a few seconds. And the question that truly matters is simple: **if no one will be used to getting their hands dirty and gaining experience anymore, who will learn to judge the quality of tasks?**

This is the heart of **co-intelligence**, the new way of working together with artificial intelligence. This topic was discussed on Friday, October 23, 2026 in Perugia, during the roundtable *"Training for co-intelligence. A paradigm shift in managerial training"*, organized as part of Confindustria Umbria's three-day training event "Digital Horizons". I came away with a clearer awareness: **the game isn't played on tools, but on our ability to judge them**.

> **In short.** With AI, producing texts, analyses and code costs almost nothing; evaluating them has become precious. The skills that matter shift toward judgment, empathy and the ability to decide in uncertainty, and the company returns to being a place where you learn every day.

## Producing costs nothing anymore. Judging does

Carlo Cordasco, philosopher at Alliance Manchester Business School, said it making the room smile: by now he could even write about quantum physics, without understanding anything about it. **Production has become free, evaluation extremely expensive**: you no longer check one document, but a thousand, each time searching for gold among the trash.

I experience this every day: I built this site largely with an AI agent, and at a certain point I could no longer reread everything. Today I decide the rules first and let thousands of automatic checks verify them. I no longer check every line: I check the criteria. It's the same lesson from [Anthropic's study on AI that builds itself](/it/editoriali/post/ai-che-costruisce-se-stessa-auto-miglioramento-ricorsivo): when the machine runs faster than we can read, we become the bottleneck.

## The expert no one is training

Today a manager knows how to evaluate because, years ago, they did that work themselves. But the young person entering the company now, if from day one it's the machine writing instead of them, when will they learn to make mistakes? And in twenty years, who will know how to recognize an error?

Cordasco again tries to explain it with a joke: when we created autopilot systems that allowed planes to fly almost by themselves, we didn't stop training pilots: we train them precisely for the moments when autopilot isn't enough. Perhaps companies will have to do the same: automate a lot, but preserve people who still know how to do things "the old way," for the day when something goes wrong.

## Doubt as a quality

The problem with today's models is that they're wrong with the same confidence with which they're right. Michele Caprio, at the University of Warwick, works to make machines learn to say how uncertain they are: not "the answer is this," but "I believe it's correct with a probability between 30 and 70%."

It's exactly what we ask of a good colleague: not that they know everything, but that they tell us when they're not sure. **The most precious quality of a machine is the same one that makes a person reliable: humility.**

This is why I refer you to the article on the new frontier of Decision AI like JEV.

## The company returns to being a workshop

In the Renaissance, you learned in a workshop, alongside a master. AI, paradoxically, brings us back there. Antonella Gandolfi, who oversees human resources at Confindustria, recalled that AI use in businesses is racing ahead, but the majority of people don't feel prepared. The answer isn't another course: it's a company where you learn every day, on the job.

With one novelty: the master isn't always the oldest. In **reciprocal mentoring**, the young person teaches the tools, the expert teaches how to read context and take responsibility. At Avanade, Monica Vetrò recounted, agent work is always rechecked by those with the experience to judge it. Because AI is a **mirror**: in a closed company it amplifies inefficiencies, in an open one it amplifies talents. And this applies to a small Umbrian business too, which today uses the same tools as San Francisco.

## The ear that hears the breakdown

Every factory has a worker who, passing by a machine, stops and says: "This one's going to break soon." It's knowledge that lives in a body, and on retirement day risks leaving with them.

In a case shared at the table, they recorded the sounds of machines along with the worker's voice explaining what they would have done. Now that story teaches the machinery, and remains for those who come after. It's the most beautiful meaning of co-intelligence: **the machine doesn't replace the worker's ear. It listens to it, and preserves it.**

## Parents' fear

In the end, a mother from the audience confessed she no longer knew which skills to recommend to her children. It was everyone's fear.

No one answered by talking about technology. Critical thinking and creativity. Letting them get bored, because ideas are born in boredom. Not letting them delegate even their own emotions to the machine. And the most surprising advice: worry less about what they study and more about who they study with, because we become similar to the people we spend time with.

After all, AI's technical limit is man himself, but that's another discussion for perhaps a separate new article.

## What remains for us

I'm a technician, and I remain convinced that without technical competence this transformation cannot be governed. But **technology helps us understand what's happening; humanity helps us decide what to do with it.** Training for co-intelligence means training judgment before tools, making the company a workshop and preserving experts' knowledge. It's the work I also do in [corporate training programs](/it/aziende).

The answer to the question I started with lies in people: in the young person teaching the older colleague, in the worker telling their trade, in the mother who never stops asking questions.

---

*From the roundtable "Training for co-intelligence," Digital Horizons, Palazzo Donini, Perugia, October 23, 2026. Moderated by Fabio Casali (SFCU Confindustria Umbria), with Monica Vetrò, Carlo Cordasco, Antonella Gandolfi and Michele Caprio. Ideas reported in my own words.*
