Three minutes.
It isn’t a metaphor or a figure of speech. Three real, measurable minutes—the kind that pass while in a classroom someone coughs, someone takes notes, someone checks the time.
During a seminar on generative AI in business, at the university, I opened Claude Cowork and gave it three CSV files of e-commerce data and a brief of a few lines. Sales, customers, marketing. Then I asked for something very “course-style”: segmentation, KPIs, ROAS by channel, charts, and a strategic report with budget recommendations.
In three minutes the AI did the analysis, built an RFM segmentation, calculated return on ad spend, produced interactive visualisations, and wrote a report that, at first glance, looked complete.
Not perfect, no. Not ready to be presented to a board without a serious review. But it was exactly the kind of deliverable that often gets requested as a final project. The kind of thing a university course schedules over a year.
I’m not writing this to belittle students or teachers. I’m writing it because that scene forced me to look in the face something that, if I’m honest, I’d rather not see: the silent divergence between the speed of technology and the speed of institutions.
Two clocks ticking different times
The university is designed to move slowly. And it isn’t necessarily a defect, on the contrary. It’s how it guarantees quality: committees, accreditations, panels, reviews, consensus. It’s almost in the DNA of these institutions not to run.
The problem is that outside the classroom, in the meantime, the world isn’t just running. It’s changing shoes every two kilometres.
A university curriculum often takes 12–24 months to be designed, approved, and implemented. Courses get planned far in advance. Textbooks have long editorial cycles, and even when they’re excellent, they arrive in a world that has already moved on.
Innovation, by contrast, thinks in releases. AI coding tools went from “curiosity” to industrial standard in a few months. AI agents complete work that previously took weeks. Startups founded by recent graduates reach huge valuations in timeframes that no longer resemble old industrial cycles.
So we end up with two clocks in the same room.
One ticks semesters and study plans.
The other ticks deploys, updates, new models, new interfaces.
The distance between the two isn’t only organisational. It’s becoming cultural.
The half-life of skills: from decades to months
There’s a concept that comes up often when we talk about work and technology: the half-life of skills. The time after which half of what you learned becomes obsolete or irrelevant.
Forty years ago that could be a decade. Today, for many technical skills, it’s around 2.5 years.
And yet you don’t even need to ask whether 2.5 years are a geological era when we talk about AI and software development. AI assistants for coding went from experiment to widespread presence in a very short time. Anyone who learned to program in 2022 without touching AI coding tools today finds themselves working with a piece of the toolkit missing.
Global estimates put even sharper numbers on the picture: by 2030, 39% of key skills required in the labour market will change. And 59% of the global workforce will need reskilling or upskilling.
Now take a bachelor’s degree: three years. Add a master’s: another two. Five years.
In the same span a student takes to complete the path, an enormous part of what the market considers “key” will have changed.
And here the question stops being theoretical. It becomes almost intimate: what are we promising the young people enrolling at university?
Entering the workforce is a wall, not a door
If it were just a content misalignment, we could discuss it calmly. The problem is that the entry-level numbers tell a harder story.
In recent years, entry-level hiring at many large tech companies has dropped sharply. In several markets we’re seeing major reductions, and in Europe in 2024 many tech positions for graduates fell significantly, with prospects that aren’t exactly rosy.
Then there are layoffs openly framed as “AI-enabled”. Companies cutting corporate roles because AI allows leaner structures. Some have publicly said AI now handles a huge share of the work.
And while this happens, among young people a feeling grows that you sense even without surveys: the idea that the market has become more closed, more selective, harder to cross.
Almost half of young job seekers believe AI has reduced the value of their university education.
It isn’t science fiction. It’s the present. And when the present knocks at your door, you can’t keep postponing the conversation.
The entry-level paradox: when AI eats the first rung
There’s a detail that makes everything more insidious.
The implicit entry-level pact, for years, was this: you do repetitive work, you cut your teeth, you learn. In exchange you get mentorship, context, growth.
That repetitive work is exactly what AI absorbs best.
So the first rung of the ladder disappears.
This is the part that worries me most, because it’s a “transmission” problem of skills. If you don’t hire juniors today, who becomes senior tomorrow? Someone described this choice as “eating the seed corn”. You save now, you pay later, and you pay with interest.
Meanwhile the university risks a vicious circle: training people for entry-level roles that are shrinking, using curricula that can become old before even reaching cruising speed.
The 2026 graduate comes out with 2023 knowledge on average, because that’s the year the curriculum was designed, for a market that has since gone through two or three revolutions.
Three minutes vs. a year, and what I saw in students’ eyes
Back to that demo, because there—more than in the charts—was the emotional part.
The brief was a few lines. Three CSVs attached. Zero configuration.
In three minutes the AI produced aggregated metrics, segmentation, ROAS, visualisations, recommendations.
Again: it wasn’t perfect. But it was good enough to set off a silent question.
And the question, in the classroom, showed in the eyes.
I didn’t see panic. I saw concentration. I saw that form of attention that arrives when you understand the ground under your feet is really moving.
I don’t know why, but that gave me hope. Maybe because it seemed the students weren’t pretending. They were looking.
And maybe that’s where the move starts: from not pretending.
Because the message, in the end, was clear. The “what” you learn is becoming less important than “how” you learn to learn.
Italy: goodwill, geological timeframes
In Italy we aren’t standing still. It would be unfair to say so.
In recent years AI training offerings have grown, with new degree courses, master’s programs, contamination with other disciplines, advanced training in specific fields. There are national strategies, university-business collaborations, e-learning platforms.
These are important steps.
But there’s a big “but”: they move at institutional speed. Tenders, committees, collegial bodies. Meanwhile outside, things move at deployment speed.
When a course gets activated after years (or even months) of process, the tools it teaches risk having already changed twice. Not because the course is poorly designed. Because time has become the adversary.
And here the divergence hurts most, because it isn’t a question of incompetence. It’s a question of architecture.
Universities guarantee quality through slowness.
Technology guarantees relevance through speed.
For years these two logics coexisted. Now they’re starting to clash.
What we’re teaching, and what maybe we should be teaching
The skills growing fastest in the coming years are a strange and beautiful mix. On one side AI and big data, cybersecurity, technological literacy. On the other creative thinking, resilience, flexibility, curiosity, lifelong learning, leadership.
Technical and deeply human together.
So I wonder if the point isn’t exactly this.
Teaching a tool isn’t enough. You have to teach how to think, adapt, communicate, decide under uncertainty. Because tools change. And they change fast.
There’s also a phenomenon I think we’re reading wrong: the widespread dependence of students on AI.
We often call it laziness. Sometimes it is, of course. But sometimes it’s a signal. Students sense when they’re being asked to invest energy in skills that look irrelevant or already obsolete. And they look for shortcuts because, from their point of view, the game isn’t worth the candle.
Universities (and schools) responding by banning AI are fighting a lost war.
Those that integrate it superficially risk doing something else: normalising the use without really changing the goal.
Maybe what’s needed is a more radical rethink of what “competence” means in 2026.
The end of the static curriculum
I don’t presume to have definitive solutions. The more I think about it, the more it seems a problem full of real trade-offs.
But some directions, at least, are visible.
The first is modularity. Abandon the idea of the monolithic multi-year curriculum and shift toward short, updateable, composable modules. Micro-credentials that get refreshed every semester. It isn’t a “modern” whim. It’s a way to realign education to a world that doesn’t wait.
The second is bringing real problems inside, not exercises. The difference is simple: an exercise has a known solution; a real problem doesn’t. AI is excellent with known solutions, and that’s exactly why it makes many traditional projects obsolete. What stays hard is choosing the right question, managing ambiguity, negotiating with stakeholders, understanding when a piece of data is “correct” but out of context.
The third is treating AI as environment, not as subject. Not an isolated course, but a transversal presence, the way the calculator didn’t kill mathematics but changed what “knowing maths” means. AI probably won’t kill education, but it’ll change what being competent means.
Then there are operational collaborations with companies. Not partnerships that produce a paper in three years, but shared work on real problems, with the same tools, in the same timeframe. It’s complex, I know. But it’s also one of the few ways to recover time.
Finally, teaching how to learn. It sounds like a poster slogan, but today it’s almost the only future-proof thing—and maybe the one the university has lost over the last 30 years. Metacognition, critical thinking, the ability to evaluate new tools, to integrate them without becoming dependent on them.
The real danger isn’t AI, it’s irrelevance
If the academy is perceived as too slow, too expensive, out of time, the risk is that people start bypassing it. Not out of malice, but out of necessity.
And here comes the most uncomfortable question, the one I fear many students are already asking themselves.
When you discover you can get in three minutes with an AI assistant what your course taught you to do in a year, you don’t think “great, I have solid foundations”. You think: why did I invest three years of my life and thousands of euros?
The answer exists, and it’s a good answer.
The value of university shouldn’t be the transfer of pointwise skills. It should be a mental framework, critical thinking, the capacity to argue, the network of relationships, exposure to different disciplines, personal maturation.
But if the curriculum is built as if the main value were “learn how to make a dashboard”, that answer sounds hollow. And it hurts to say, because behind many courses there’s real dedication.
The urgency to act, without panic
Globally we invest very little, in proportion, in adult lifelong learning. And yet increasing that investment could generate enormous value by 2030.
It isn’t only an educational problem. It’s economic. It’s social.
Universities have an advantage no AI platform has: they can teach how to think critically, how to collaborate deeply, how to navigate ethical ambiguity.
But only if they stop competing on terrain where AI wins—the transmission of information and the execution of structured tasks—and concentrate on what makes them irreplaceable.
The time for this transition isn’t “the next few years”. It’s now.
Every semester that passes with unchanged curricula is a semester of students coming out less prepared than they could be.
The speed of AI won’t slow down to wait for educational committees.
That three-minute demo wasn’t a provocation. It was a mirror.
And the thing that struck me most wasn’t the speed of AI. It was the slowness of everything else.
Key takeaways
The half-life of technical skills has dropped below the duration of a master’s degree.
If AI eats entry-level work, the channel through which tomorrow’s seniors are formed disappears too.
The university stays irreplaceable only if it stops competing on terrain where AI wins.
Questions & answers
What does it mean that an AI agent replicates a year-long university project in 3 minutes?
That a good portion of what a CS course uses as “didactic project” (implementing a data structure, a small framework, a tool) is now producible by an agent in a few minutes. This doesn’t devalue learning—but it forces us to ask what the university is actually teaching. If the value was the practice of writing code, that practice has dematerialised.
So is the university irrelevant?
No, but it has to choose its side. It can stay central if it shifts focus from products (code, tools) to processes (how you reason about a problem, how you design a system, how you evaluate an alternative). It can become irrelevant if it keeps assigning as a year’s work what an agent produces in an afternoon—because the student learns to hide it, not to understand it.
How should curricula change?
Evaluate understanding, not the final product. Oral exams on explaining the code (generated or not), projects with constraints AI can’t solve alone (niche domains, proprietary data, motivated trade-off choices), reduction of productive assignments in favour of design, critique, and debugging exercises. The effective university after AI is more vertically demanding, not more permissive.
What can a student do today not to be crushed by the transformation?
Use AI, declare it, and spend the freed time on understanding. The graduate who pretends not to use agents because it isn’t allowed, and the graduate who uses them as a shortcut without understanding, end up the same—with a piece of paper and no skills. Anyone who instead uses agents to accelerate the mechanical part and invests the time in interrogating the results comes out with a rare and very requested combination.
The author
