AI is already in your future

 

AI adoption in Europe and Italy since 2022, and what it means for your career

Lecture delivered 20th of October 2026

Lecture for the undergraduate degree in Artificial Intelligence Management

Duration: about 60 minutes 50 of lecture and 10 of discussion)

Short link: https://bit.ly/OPTI_class20ottobre2026

Slides: https://app.decktopus.com/nano/6094779/present?code=LktCyw

Future of work Italian version: https://bit.ly/DrAlbert_futurodellevoro

Jensen Huang interview: https://bit.ly/interview_huang

Learning objectives

By the end of the lecture you will be able to:

• explain why generative AI was adopted first by individuals and only later by organizations;
• read European and Italian AI adoption data critically, distinguishing between use and value created;
• place Italy’s position in a historical and European perspective;
• use a structured framework to assess the return on investment (ROI) of AI under conditions of high uncertainty;
• discuss the advantages and limits of the European AI Act;
• define a concrete plan to develop AI skills and demonstrate them to an employer.

Outline

Part	Topic	Time
1	Introduction: a technology you already have in your pocket	5 min
2	Four facts about generative AI	12 min
3	Europe since 2022: where does Italy stand?	10 min
4	Italy: rapid growth, large gaps, little measurement	10 min
5	Why Europe is lagging	7 min
6	Your role in the AI economy	6 min
7	Conclusion and a two-week action	2 min
Q&A	Questions and discussion	10 min

Minimal glossary

Before we begin, let us fix some terms we will use often.

• Generative AI: systems that produce new content (text, images, code, audio) from an instruction given by the user, called a prompt [2].
• LLM (Large Language Model): a large-scale language model, such as ChatGPT, Gemini, Claude or DeepSeek.
• Benchmark: a standardized test used to compare the performance of models.
• AI agent: a system that plans, uses external tools and carries out multi-step tasks with a degree of autonomy [8].
• Open-weight model: a model whose “weights” (the learned parameters) are published and can be downloaded, run locally and modified [14].
• API: an interface that lets other software use a paid model without owning it.
• SMEs: small and medium-sized enterprises, meaning those with 10–249 employees in Istat statistics [17].
  
  

Part 1. Introduction: a technology you already have in your pocket

The concept. Every major technology is adopted in two phases: first by individuals, then by institutions. Who benefits from a technological revolution depends largely on how long the interval between these two phases lasts. Today, generative AI sits exactly in that interval.

Classroom moment, poll. Answer in the chat:

• When did you first use an LLM? In 2022, in 2023, last year, or never?
• Have you used one this week, to summarize a text, translate an email or fix your CV?

Almost all of you probably answered yes to the second question. According to Stanford University’s AI Index 2026, four out of five university students use generative AI. The same report calculates that generative AI reached 53% adoption in the population within three years of launch, faster than the personal computer and the internet [1]. Your parents took years to get used to email. You took a few months to get used to a machine that answers you in writing.

Why it matters. The question is not whether AI will change your working life, but what place you will occupy when it happens. The title of the lecture should be taken literally: AI is “hiring”, in the sense that companies are already looking for people who can work with these tools.

Part 2. Four facts about generative AI

Many ideas circulate about this technology, but some basic facts are often misunderstood, even by experienced managers. Let us look at four.

2.1 Extremely fast adoption, but mostly personal

The concept. Technologies usually enter from the top: the company buys the computers and then employees learn to use them. With generative AI the opposite happened: it came in from the bottom, through individuals.

The data. According to Eurostat, in 2025, 32.7% of Europeans aged 16 to 74 used generative AI tools. Personal use (25.1%) far exceeds use at work (15.1%) and use for formal education (9.4%) [2]. For us the national figure is worrying: at 19.9%, Italy is second to last in the EU. Only Romania does worse (17.8%), and Italy is very far from Denmark (48.4%) [2].

The figure for young people is even more striking. Among those aged 16 to 24, 63.8% of young Europeans use generative AI, but only 47.2% of young Italians. In this age group too, Italy is second to last, ahead only of Romania (44.1%) [3]. In Europe the majority of young people use these tools; in Italy, fewer than half.

When AI is used at work, it is often the employee who brings the tool. In Microsoft and LinkedIn’s 2024 global survey, 75% of knowledge workers said they used generative AI at work, and 78% of them used their own tools [4]. Italy follows the same pattern: 47% of workers use AI at work, but only 19% use exclusively tools provided by the company [5]. The technology, in short, came in through the back door.

Why it matters for a manager. When you enter an organization, you may know these tools better than the people who lead you. A recent graduate is rarely in this position, and it is an opportunity. For the company, however, it is also a governance risk, as we will see when we discuss “shadow AI” (§4.5).

2.2 Models improve drastically, and often

The concept. Any judgment about a technology that changes every month has a short expiry date. If you tried AI a year ago and found it unreliable, you judged a product that no longer exists today.

The data. Take a very difficult benchmark, GPQA. It contains 448 multiple-choice questions written by experts in biology, physics and chemistry. Experts who hold or are pursuing a doctorate in the subject answer correctly 65% of the time (74% if one excludes the obvious mistakes they themselves recognized). Highly skilled non-experts reach only 34%, even though they spend on average more than 30 minutes on each question with free access to the web. This is why the questions are called “Google-proof”. When the test was published, in November 2023, the best GPT-4-based system scored 39% [6]. In 2025, average accuracy on the hardest subset, GPQA Diamond, reached 93%, twelve points above the experts’ reference threshold (81.2%) [1].

Let us do the calculation together. Going from 39% to 93% is an increase of about 138%: the score more than doubled in about 25 months. On average, that is a compound improvement of about 3.5% per month. My own comparisons indicate that some Chinese open-weight models, such as Moonshot AI’s Kimi family, have improved even faster.

Two caveats, because a good analyst always accompanies data with their limitations:

• Ceiling effect. Near the maximum, each additional point is harder to gain and says less. Stanford found invalid questions in widely used benchmarks, in proportions ranging from 2% to 42% [1].
• Benchmarks are not jobs. The AI Index 2026 speaks of a “jagged frontier”: a model that wins gold at the International Mathematical Olympiad reads an analog clock correctly only about half the time [1].

Costs have fallen even faster than capabilities have grown. The price of using a GPT-3.5-level model fell by more than 280 times between November 2022 and October 2024 [7]. This explains why in 2026 Jensen Huang, CEO of Nvidia, changed his tone. In February he stated that agentic AI had reached “an inflection point” in the previous two or three months [8]. In June, at the Microsoft developer conference, he said: “All of a sudden, because of agentic systems... AI is now useful” [9]. Even an enthusiast of the industry thus admits, in effect, that the first years of generative AI were more impressive than useful.

Why it matters. For a manager, technical knowledge about AI has a very short “expiry date”. What counts is the ability to regularly re-test what the technology can do.

2.3 Rapid improvement creates an incentive to wait

The concept. If a technology improves every month, waiting seems rational. This reasoning is half right, and the wrong half can be costly.

A system built on today’s model can be superseded in six months. For a small business with limited budgets, being a “fast follower” may seem wiser than being the first to adopt. Here, however, a fundamental concept for AI management comes in: the “productivity J-curve”, described by the economists Erik Brynjolfsson, Daniel Rock and Chad Syverson [10].

Key concept: the J-curve. With general-purpose technologies (electricity, computers, now AI), the gains do not come from buying the technology. They come from intangible investments: reorganizing processes, retraining people, redesigning roles. At first these investments cost money, and measured productivity may even fall before it rises. The graph looks like a J.

These intangible investments cannot be bought ready-made on the day the technology “matures”: they must be built over time. In an analysis of AI ROI I called this phenomenon breakthrough volatility: a single architectural leap can make ROI calculations already done obsolete overnight. It is an argument for caution on large infrastructure commitments [11].

Why it matters. The balanced conclusion is this: on infrastructure it may make sense to wait, on skills never. This applies to companies and it applies to you.

2.4 Two ecosystems, and Europe should use both

The concept. The AI market is often described as a race between companies. It is more useful to see it as a competition between two business models, closed and open, which for historical reasons are concentrated in two countries.

• US commercial models (OpenAI, Google, Anthropic and others): mostly closed. You pay for access through an app or an API.
• Many Chinese models (DeepSeek, Alibaba’s Qwen, Moonshot’s Kimi): open-weight. The model can be downloaded, run on your own servers and modified [14].

The gap has narrowed but has not disappeared. In 2024 it had shrunk from 8% to 1.7% on some benchmarks [7]; it then reopened, and in March 2026 the best closed model outperformed the best open model by about 3.3% [1]. According to Epoch AI, in cybersecurity-related computing capabilities, open-weight models trail closed ones by six to seven months [12].

Why do open models seem to improve faster? Part of the answer lies in a concept from economic history that we will meet again when we discuss Italy: Alexander Gerschenkron’s “advantages of backwardness”. Catching up costs less than leading, because the follower can copy solutions already tested by others [15].

Price matters even more. As reported by Prof G Media and summarized on my blog AI4TL, the price per million output tokens was about 45 dollars for OpenAI’s flagship model and 50 for Anthropic’s, against 0.87 dollars for DeepSeek. According to the same sources, the share of global AI traffic served by free Chinese models would have gone from less than a third to about two thirds [13]. If intelligence becomes a cheap, open commodity, the advantage passes from whoever builds the smartest model to whoever applies it fastest [14].

Why it matters for a European manager. The sensible strategy is to use both ecosystems: US models where they are clearly better; open models, Chinese or European such as those of the French company Mistral AI, where cost, privacy or customization matter. There is one key condition: an open-weight model can run on European servers, under European law, without sending customer data abroad. This is a choice of data sovereignty, not only of price.

Part 3. Europe since 2022: where does Italy stand?

3.1 The numbers

The concept. Averages hide more than they reveal. To understand Italy’s position it is not enough to look at the ranking: one must look at the speed at which it is moving and where the gaps lie.

Every year Eurostat measures how many European enterprises with at least ten employees use AI. In 2025, 19.95% used at least one AI technology, 6.47 percentage points more than in 2024 [16]. Denmark leads (42.03%), followed by Finland (37.82%) and Sweden (35.04%); at the bottom are Romania (5.21%), Poland (8.36%) and Bulgaria (8.55%) [16]. Firm size matters more than geography: 55.03% of large enterprises use AI, 30.36% of medium-sized ones and 17% of small ones [16].

Classroom moment. Before seeing the Italian figure, write your estimate in the chat: is Italy closer to Denmark or to Romania?

According to Istat, Italy stands at 16.4%: below the European average, but growing strongly from 5.0% in 2023 and 8.2% in 2024 [17]. In a single year adoption doubled. This is rapid growth. But the European “Digital Decade” target envisages 75% of enterprises using AI by 2030, and Italy has covered only 21.9% of it [17].

Why it matters. A manager must read level and trend together. Italy is behind, but it is catching up fast. To understand why it started so far behind, we need a little history.

3.2 A long view: four industrial revolutions

The concept. Economic historians use the idea of successive industrial revolutions to describe how new energy sources and new technologies reorganize entire societies, not just factories.

Klaus Schwab, founder of the World Economic Forum, describes four [18]:

• The first (late 18th century): steam, coal, mechanized textiles. It begins in Great Britain.
• The second (from about 1870): electricity, steel, chemicals, mass production. It is led by Germany and the United States.
• The third (from the 1960s): electronics, computers, the internet.
• The fourth: the fusion of physical, digital and biological systems.

Generative AI is the most visible face of the fourth. Earlier revolutions multiplied what our muscles could do; this one multiplies what our minds can do. As I wrote on the AI4TL blog, for much of the last century technological progress coincided with progress in hardware. That era is closing, and the center of gravity is shifting toward software and models [19].

3.3 Italian industrial history: late, then brilliant

The concept. Italian history shows that starting late does not condemn a country to stay behind. What matters is the ability to build the institutions that allow innovations to scale.

Italy arrived late to the first two industrial revolutions, for reasons documented by Vera Zamagni [20]. It had almost no coal, and later bet on the hydroelectric power of the Alps, the “white coal”. At Unification, in 1861, the majority of Italians could not read or write. Its industrialization, concentrated in the Milan–Turin–Genoa triangle, is a textbook case of late development in Gerschenkron’s sense: banks and the State played the role that private entrepreneurs had played elsewhere [15].

In the third revolution Italy was, for a moment, at the world frontier. Under Adriano Olivetti, the Ivrea company built the Elea 9003 (1959), one of the first fully transistorized commercial computers, designed by a group led by the engineer Mario Tchou. In 1965 it launched the Programma 101, often considered the first commercially successful desktop computer. But Adriano Olivetti died in 1960 and Tchou in 1961, and the electronics division was sold in 1964. Lorenzo Soria called it a lost opportunity [21].

Why it matters. The lesson for management students is clear: Italy never lacked talent; it often lacked the ecosystem to scale its own innovations.

3.4 Everyday obstacles: skills, language, talent

The concept. Advanced technologies rest on foundations that are anything but advanced: basic digital skills, a shared working language, administrative systems that are easy to use.

• Digital skills. According to the European Commission’s 2025 Digital Decade report, only 45.8% of Italians have at least basic digital skills, far from the EU target of 80% by 2030. The gaps are wide among people with lower levels of education, but also among young people [22].
• Language. The AI Index 2026 notes that AI works best in English, and that in other languages the performance gaps are wider than global benchmarks suggest [1]. This is why Italian-specific evaluation tools matter, such as ITALIC, a “culturally aware” benchmark of 10,000 questions across 12 domains [1]. And this is why good English remains, for you, a career multiplier.
• Talent. The AI Index counts, in Italy, 22.43 leading AI authors and inventors per 100,000 inhabitants, against 110.45 in Switzerland and 58.10 in Germany. This is the lowest value among the 15 countries in the chart [1].

3.5 Industry 4.0 and who leads this revolution

The concept. Industrial policy can get machinery bought quickly. Changing the way a company thinks takes much longer.

In Italy the expression “fourth industrial revolution” spread with the National Industry 4.0 Plan of 2016. Its tax incentives, including the hyper-depreciation allowance, rewarded investment in connected machinery [23]. The plan later became Transition 4.0 and then 5.0. These instruments worked: in 2025, 79.5% of Italian SMEs had at least a “basic” level of digitalization [17]. Many firms, however, experienced Industry 4.0 as a tax opportunity rather than as a change in how they work.

Unlike railways or electrification, the generative AI wave is driven mainly by private companies: in 2025 industry produced more than 90% of the relevant models [1]. Mariana Mazzucato reminds us, however, that the internet, GPS and much of the research underlying neural networks were funded by taxpayers [24]. Public investment in education and research remains decisive. As I wrote on the blog Life is a Journey of Learning, programs funded by the PNRR (Italy’s National Recovery and Resilience Plan) could transform Italy from a country of brain drain into a “skills engine”, provided that universities and students seize the opportunity [25]. The Politecnico di Milano warns that the end of PNRR funds, in the absence of a strategic plan to replace them, risks undoing this progress [5].

Part 4. Italy: rapid growth, large gaps, little measurement

4.1 The small-business gap

The concept. The Italian economy rests on small family firms. A technology that rewards scale hits Italy harder than countries dominated by large groups.

Istat data [17]:

• 53.1% of large Italian enterprises use AI (it was 32.5% in 2024);
• among SMEs the share has doubled, but only from 7.7% to 15.7%;
• a large enterprise is now more than three times as likely as an SME to use AI;
• the gap is widening: about 20 percentage points in 2023, 25 in 2024, 37 in 2025. Over the same period almost all other digital gaps between large and small enterprises narrowed;
• the North-West leads, up from 8.9% to 19.3%.

The Politecnico di Milano reports different figures: 71% of large enterprises and only 8% of SMEs. The difference depends on what is measured: the Politecnico counts AI projects, that is, tailored solutions that use company data, not the mere use of a tool [5]. The direction, however, is the same. The global picture also matches: according to McKinsey’s State of AI 2026 report, the share of large organizations scaling AI agents rose from 27% to 40% in a year, while among smaller firms it stayed flat at 22% [30].

A methodological note a manager must know. These surveys exclude enterprises with fewer than ten employees, who are the great majority of Italian enterprises. The real gap is therefore probably wider than the measured one.

Why it matters for you. There is a positive side. Hundreds of thousands of SMEs will need people able to bring AI into marketing, accounting, logistics and design. The task is not to build models but to apply them. This could be the description of your first job.

4.2 The market and the +93% in job postings

The concept. The clearest sign that a technology has left the enthusiasm phase is when it enters job descriptions.

In 2025 the Italian AI market reached 1.8 billion euros, growing by 50%; 46% of the total comes from generative AI solutions or hybrid projects [5]. The Politecnico, with Lightcast, analyzed 3.2 million online job postings published in Italy: about 44,000 required AI skills, 93% more than a year earlier. These skills appear in 76% of postings for highly qualified white-collar profiles, in 27% of those for Chief HR Officers and in 12% of those for Chief Marketing Officers [5]. The 44,000 postings are about 1.4% of the total, a figure consistent with Stanford’s estimate that AI skills are required in 1.33% of Italian postings, against 3.31% in Spain [1].

Why it matters. “AI skills” are no longer a requirement reserved for engineers. They are becoming what “good knowledge of Excel” or “fluent English” were for earlier generations.

4.3 What is not measured: productive use

The concept. Statistics steer policy, and what they leave out can matter as much as what they include. Official surveys record whether a firm uses AI, not whether that use creates value.

Istat offers a revealing clue. In 2025, 33.4% of Italian enterprises using AI did not indicate any business area in which they use it, against 15.5% in 2024; 83.3% of these are small firms [17]. Istat speaks of adoption that is “increasingly widespread but still poorly structured”, at an early and experimental stage [17]. Among large enterprises using generative AI, only 54% try to measure its benefits, almost always by collecting employees’ opinions, and only 11% do so periodically and in a structured way [5].

International research explains why measurement is decisive. MIT’s NANDA project reported that 95% of organizations see no measurable bottom-line impact from generative AI pilot projects, despite corporate investments of between 30 and 40 billion dollars [26]. Controlled studies on specific tasks, by contrast, show notable gains:

• Customer support: an AI assistant raised productivity by 14% on average, and by much more for less experienced agents [27].
• Consulting: Boston Consulting Group consultants using GPT-4 performed better on tasks within the AI’s capabilities, but worse on a task just outside them. The authors call this the “jagged technological frontier” [28].
• Professional writing: ChatGPT cut time by about 40% and improved quality [29].

Italian workers confirm this. 39% say AI saved them more than 30 minutes in their last two tasks using it. Even more interesting: 41% say that thanks to AI they carry out activities they otherwise could not do, or could only partly do [5].

Key concept. The value of AI is created at the level of the individual task but lost at the level of the firm. Closing this gap is management work, and a recent graduate can do it.

4.4 The ROI puzzle: two filters and three lenses

The concept. Return on investment (ROI) is the language in which boards decide. The classic calculation presupposes stable costs, benefits and timing, and AI guarantees none of the three. The solution is not to give up measuring, but to measure with more discipline.

About one in three large Italian enterprises struggles to estimate the cost-benefit ratio in advance [5]. In a recent working framework I proposed that, before any calculation, two filters be applied [11]:

• Strategic fit: what objective does the project serve? Speeding up production, lightening workload, improving quality or reducing development costs?
• Task suitability: can AI perform that task reliably, only in part, or not yet? Because the frontier moves every month, the check must be repeated periodically.

Only projects that pass both filters deserve analysis with three lenses, used in sequence [11]:

• Process cost comparison (efficiency). A before/after comparison. In one case, a manufacturing firm that automated invoice handling cut annual costs from 30,080 to 11,632 euros. With an integration cost of 8,000 euros, ROI was 53% in the first year and 159% in the second, with payback in seven months.
• Productivity multiplier (growth). It measures the value of the capacity freed up. A professional services firm cut research time by 35%, allowing ten consultants to handle 80 more projects: 272,000 euros of additional gross margin and a first-year ROI of 71%. When AI makes possible a service that did not exist before, the opportunity cost is low.
• Risk-adjusted NPV (strategy). A net present value over 12 to 36 months with probability weights, for example 20% success, 30% partial success and 50% failure. It allows an AI project to be compared with non-AI investments.

The framework recommends recalculating ROI every six months and treating data cleaning as a major cost from the outset [11].

Why it matters. Uncertainty is not an excuse to give up rigor. Those who learn to measure early, even roughly, will be the first to understand which projects deserve to grow.

4.5 Shadow AI: hidden adoption

The concept. Rules always arrive after behavior. We speak of “shadow AI” when employees use AI tools without the company knowing or approving it. It is the distance between what the organization allows and what people actually do.

According to MIT, workers at more than 90% of companies use personal AI tools, while only 40% of companies have official LLM subscriptions [26]. In Italy only 19% of those who use AI do so exclusively with company tools. Among large enterprises, 24% ban generative AI tools not provided by the organization, and only four in ten have defined usage guidelines [5]. My hypothesis is that in small firms hidden use is even more intense: they rarely have rules or an IT department, but every employee has a smartphone. If so, official statistics underestimate the real use of AI in Italian SMEs.

The risks are concrete: confidential data pasted into public tools, uneven quality, unclear accountability. It is, however, a transitional problem: when AI is built into commonly used software, hidden use will become normal use.

Why it matters for a manager. The company’s job is to provide safe tools, simple rules and training. Yours is to be the employee who uses AI transparently, responsibly and competently.

Part 5. Why Europe is lagging

5.1 Draghi’s diagnosis

The concept. An economy’s competitiveness depends less on any single invention than on the ability to turn inventions into growing companies.

In September 2024 Mario Draghi presented to the European Commission the report The Future of European Competitiveness [31]. His diagnosis in brief:

• Europe missed the digital revolution and has few large technology companies;
• its markets are fragmented by national rules and separate capital markets;
• promising start-ups move to the United States to find funding;
• Europe is stuck in a “middle technology trap”: it excels in mature sectors such as automobiles and invests little in disruptive technologies.

The report contains 383 recommendations and estimates a need for additional investment of 750–800 billion euros per year, equal to about 4.4–4.7% of EU GDP [31].

The size of the gap is striking. In 2025 private investment in AI reached 285.9 billion dollars in the United States and 20.92 billion in the whole of Europe [1]: the United States invests almost fourteen times more than Europe.

5.2 The implementation gap

The concept. Making a diagnosis is easy; putting it into practice is hard, especially when 27 governments must agree.

The Commission responded with the Competitiveness Compass of January 2025 [32] and with the AI Continent Action Plan, which provides for “AI factories” and “gigafactories”, large computing centers to train and run models [33]. Italy hosts one of these AI factories, at CINECA in Bologna. Implementation, however, is slow. One year after the report, according to the European Policy Innovation Council, only 11.2% of the 383 recommendations had been fully implemented; by July 2026 it had reached 15.7%, and the recommendations on digital and research were among the slowest [34].

5.3 The AI Act: shield or brake?

The concept. Every rule is a compromise between protection and freedom to experiment. The real question is not whether to regulate AI, but whether Europe has found the right balance.

The AI Act (EU Regulation 2024/1689), in force since 1 August 2024, classifies AI systems by risk [35]:

Risk level	Example	Regime
Unacceptable	social scoring	banned since February 2025
High	recruitment, education, credit, critical infrastructure	strict obligations
Limited	chatbots	transparency obligations
Minimal	spam filters	essentially free

In November 2025 the Commission proposed a “Digital Omnibus” package, justifying it by the lack of technical standards and delays in appointing national authorities [36]. With the agreement of May 2026, the rules for high-risk systems will apply from 2 December 2027 for stand-alone systems and from 2 August 2028 for those embedded in products [37]. With Law 132/2025 Italy became the first Member State to adopt a national AI law alongside the AI Act [38], [1].

Classroom moment, lightning debate. Split into two groups: one argues that the AI Act is a shield, the other that it is a brake. Two minutes per group.

Arguments for the shield:

• it protects citizens from discriminatory selection algorithms, deepfakes and surveillance;
• it gives businesses a single rule for 27 markets;
• through the so-called “Brussels effect” described by Anu Bradford, European rules often become global standards [39];
• across the 25 countries surveyed, on average 53% of people trust the EU as an AI regulator, against 37% for the United States and 27% for China [1];
• as I discussed on AI4TL, according to Barack Obama voluntary self-restraint by the labs is a temporary remedy, not a solution [40].

Arguments for the brake:

• compliance costs weigh mostly on SMEs, that is, on precisely the Italian productive structure;
• among Italian firms that assessed AI without then adopting it, 47.3% cite lack of clarity on legal consequences, against 40.1% in 2023 [17]; in the EU the share is 52.52% [16];
• Draghi warns that Europe regulates technologies it does not produce [31];
• the postponements themselves suggest that the initial timetable was too ambitious [37].

My position. The AI Act is a necessary framework, but it is not a sufficient strategy. Regulation can make AI trustworthy; only investment, integrated markets and skills can make Europe competitive.

Why it concerns you directly. Since February 2025, Article 4 of the AI Act requires organizations that use AI to ensure that their staff have a sufficient level of AI literacy [35]. Your future employers are legally obliged to care about how well you understand AI. More than half of large Italian enterprises have already launched literacy initiatives, but only 15% have a structured compliance project [5].

Part 6. Your role in the AI economy

6.1 What companies mean by “AI skills”

The concept. A skill counts only if you can describe it precisely. Today “AI skills” is one of the vaguest expressions in job postings.

The World Economic Forum’s Future of Jobs Report 2025, based on more than 1,000 employers, forecasts that by 2030 170 million jobs will be created and 92 million eliminated, for a net positive balance of 78 million. It also forecasts that 39% of workers’ core skills will change. AI and big data top the list of fastest-growing skills, together with creative thinking, resilience and lifelong learning [41]. 63% of employers name the skills gap as the main obstacle to transformation [41]. IBM points to a revealing discrepancy: 86% of CEOs think their employees know how to collaborate with AI, but only 25% of workers use it regularly [42].

In practice, the expression covers three levels:

• AI literacy: knowing what the tools can and cannot do, including “hallucinations” (confidently invented facts), biases and privacy risks.
• Applied use: knowing how to write good instructions, verify results, combine AI with spreadsheets and documents and, increasingly, supervise AI agents. IBM stresses the importance of defining agents’ goals and limits, monitoring their actions and knowing when human approval is needed [42].
• Technical skills: data science, programming, machine learning and putting models into production. These are needed by a narrower group of specialists.

Almost none of you needs the third level; all of you need the first and the second. Companies that use AI to empower people rather than replace them tend to build stronger organizations [43]. Huang himself says that being surrounded by “superhuman” capabilities gives him the strength to take on “more and more ambitious things” [43]. Remember the jagged frontier [28]: the most valuable skill is judgment, and judgment comes from knowing your own field.

A warning for recent graduates. In the United States, employment of software developers aged 22 to 25 has fallen by almost 20% compared with 2024, while that of older colleagues continued to grow [1]. Junior tasks are the easiest to automate. Your advantage must be what was rarely asked of a junior: knowing how to direct, check and integrate the work of AI.

6.2 How to show an employer that you have these skills

Writing “AI skills” on your CV proves nothing. Evidence of results does.

• Build a small portfolio. Solve two or three real problems with AI, taken from your studies, a part-time job or the family business. Document what you tried, what did not work and how you verified the results.
• Earn a micro-certification. Elements of AI from the University of Helsinki is free and available in Italian [45].
• Show responsible use. Explain that you verify sources, protect confidential data and know the basics of the AI Act. A company worried about shadow AI will notice.
• Talk about results, not tools. “I cut the preparation of the weekly report from three hours to forty minutes” is worth far more than “I use ChatGPT”.
• Practice your English. It doubles the performance of AI tools.

This is the principle of learning by doing. On my blog, together with two students, we argued that Italian schooling teaches business economics but rarely gives students the chance to do business [44]. AI is learned the same way.

6.3 What this means for your study choices

Choosing a path to escape AI is a losing strategy; choosing one that AI can multiply is a winning one. Study what you are genuinely passionate about, because it is deep knowledge that lets you tell whether an AI result is correct. Then add data literacy, applied use of AI and English. Think of AI as a multiplier of your discipline:

• the economist who automates data analysis;
• the languages graduate who supervises machine translation;
• the tourism graduate who designs personalized itineraries;
• the engineer who knows both the machines and the models that monitor them.

For a small Italian business, a historian or a jurist who uses AI well can be worth more than a mediocre programmer.

6.4 Exercise: AI in your field (3 minutes)

Classroom moment. Complete this sentence in the chat: “In my field of study, AI could be used to ______.” Be specific: not “to help with research”, but “to compare 200 eviction rulings and identify recurring patterns”.

Then answer three questions:

• What does a person still have to do?
• What could go wrong?
• How would you verify the result?

Every good answer involves both the machine and the person. The machine brings volume and speed; you bring the question, the context, the judgment and the responsibility.

Part 7. Conclusion: your action for the next two weeks

The concept. Technological revolutions reward those who learn early, not those who wait for certainty.

Let us recap. Generative AI is the fastest-adopted technology in history, but mostly by individuals. It improves every month. Europe is behind the United States and China, and Italy is behind Europe: one Italian adult in five uses generative AI, against one in three in the EU [2]. Yet in one year Italy doubled adoption among enterprises [17] and almost doubled the demand for AI skills [5]. With Olivetti, Italy has already stood at the frontier. Today the biggest gap, and the biggest opportunity, lies in hundreds of thousands of small firms that experiment without measuring.

Write down a single action to complete within two weeks:

• complete the first chapter of Elements of AI in Italian [45];
• solve a real task from your course with AI and write half a page on what worked;
• compare the answers of a closed US model and an open model to the same question from your field;
• read the summary of the Draghi report and decide whether you agree;
• ask a small business owner, perhaps a relative, how they use AI and whether they would need help.

Put a date next to the action and, if you wish, share it with me. Life is a journey of learning. What will distinguish those who benefit from this revolution from those who are left behind will be neither the degree nor talent, but the choice to start learning now, and not to stop.

Review and discussion questions

1. Why was generative AI adopted first by individuals rather than organizations? What risks does this sequence create for a firm?
2. Explain the productivity J-curve. Why might an SME see its productivity worsen immediately after introducing AI?
3. The GPQA score went from 39% to 93%. What two caveats must be considered before drawing conclusions about the world of work?
4. Why do Istat and Politecnico data on large Italian enterprises differ (53.1% against 71%)? What does this teach about reading statistics?
5. Apply the two filters and three lenses to an AI project in a company you know.
6. Is the AI Act a shield or a brake for Italian SMEs? Argue using at least two data points from the lecture.

Bibliography

[1] Sajadieh, S., Fattorini, L., Perrault, R., Gil, Y., Parli, V., et al. (2026). The AI Index 2026 annual report. AI Index Steering Committee, Institute for Human-Centered AI, Stanford University. https://hai.stanford.edu/ai-index/2026-ai-index-report

[2] Eurostat. (2025, 16 December). 32.7% of EU people used generative AI tools in 2025 [Press release]. European Commission. https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251216-3

[3] Eurostat. (2026, 10 February). 64% of 16–24-year-olds used AI in 2025 [Press release]. European Commission. https://ec.europa.eu/eurostat/web/products-eurostat-news/w/edn-20260210-1

[4] Microsoft & LinkedIn. (2024, 8 May). 2024 Work Trend Index annual report: AI at work is here. Now comes the hard part. Microsoft WorkLab. https://www.microsoft.com/en-us/worklab/work-trend-index

[5] Osservatorio Artificial Intelligence. (2026, 5 February). Artificial Intelligence: adozione, trasformazione, equilibrio [Executive summary]. School of Management and DEIB, Politecnico di Milano. https://www.osservatori.net

[6] Rein, D., Hou, B. L., Stickland, A. C., Petty, J., Pang, R. Y., Dirani, J., Michael, J., & Bowman, S. R. (2023). GPQA: A graduate-level Google-proof Q&A benchmark (arXiv:2311.12022). arXiv. https://arxiv.org/abs/2311.12022

[7] Stanford Institute for Human-Centered Artificial Intelligence. (2025). The 2025 AI Index report. Stanford University. https://hai.stanford.edu/ai-index/2025-ai-index-report

[8] Stimac, B. (2026, 26 February). Nvidia’s Jensen Huang says agentic AI has arrived at an “inflection point.” CNET. https://www.cnet.com/tech/services-and-software/nvidias-jensen-huang-agentic-ai-inflection-point/

[9] Moore, M. (2026, 2 June). “AI is now useful”: Nvidia CEO Jensen Huang thinks a new era for AI is here. TechRadar Pro. https://www.techradar.com/pro/ai-is-now-useful-nvidia-ceo-jensen-huang-thinks-a-new-era-for-ai-is-here-and-its-partnership-with-microsoft-could-be-key-for-achieving-it

[10] Brynjolfsson, E., Rock, D., & Syverson, C. (2021). The productivity J-curve: How intangibles complement general purpose technologies. American Economic Journal: Macroeconomics, 13(1), 333–372. https://doi.org/10.1257/mac.20180386

[11] Schram, A. (2026, 28 May). Strategic frameworks for calculating ROI of AI investments [Unpublished manuscript].

[12] Epoch AI. (2026, 16 October). Open-weight models trail closed models in cyber capabilities by 6–7 months [Data insight]. https://epoch.ai

[13] Schram, A. (2026, 26 July). Bets, debts, and bedrooms: The week the AI economy showed its cracks. AI4TL – Artificial Intelligence for Teaching & Learning. https://ai4tl.blogspot.com/

[14] Schram, A. (2026, 26 July). Downloads, breakouts, and golden tickets: Three stories from Moonshots that explain where the world is heading. AI4TL – Artificial Intelligence for Teaching & Learning. https://ai4tl.blogspot.com/

[15] Gerschenkron, A. (1962). Economic backwardness in historical perspective: A book of essays. Belknap Press of Harvard University Press.

[16] Eurostat. (2025, December). Use of artificial intelligence in enterprises. Statistics Explained. https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises

[17] Istat. (2025, 15 December). Imprese e ICT – Anno 2025 [Statistical report]. Istituto Nazionale di Statistica. https://www.istat.it

[18] Schwab, K. (2016). The fourth industrial revolution. World Economic Forum.

[19] Schram, A. (2026, 1 September). A brief history of technology: From the vacuum tube to the end of Moore’s Law. AI4TL – Artificial Intelligence for Teaching & Learning. https://ai4tl.blogspot.com/

[20] Zamagni, V. (1993). The economic history of Italy, 1860–1990. Clarendon Press.

[21] Soria, L. (1979). Informatica: un’occasione perduta. La Divisione elettronica dell’Olivetti nei primi anni del centrosinistra. Einaudi.

[22] European Commission. (2025). Italy 2025 Digital Decade country report. https://digital-strategy.ec.europa.eu/en/factpages/italy-2025-digital-decade-country-report

[23] Ministero dello Sviluppo Economico. (2016). Piano nazionale Industria 4.0. Italian Government.

[24] Mazzucato, M. (2013). The entrepreneurial state: Debunking public vs. private sector myths. Anthem Press.

[25] Schram, A. (2025, 27 December). Italy’s education revolution: From brain drain to skills engine by 2026. Life is a Journey of Learning. https://albertschram.blogspot.com/2025/12/italys-education-revolution-from-brain.html

[26] Challapally, A., Pease, C., Raskar, R., & Chari, P. (2025). The GenAI divide: State of AI in business 2025. MIT NANDA.

[27] Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044

[28] Dell’Acqua, F., McFowland, E., III, Mollick, E. R., Lifshitz-Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality (Working Paper No. 24-013). Harvard Business School.

[29] Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192. https://doi.org/10.1126/science.adh2586

[30] Schram, A. (2026, 11 October). AI in business in Italy: From pilots to profit. AI4TL – Artificial Intelligence for Teaching & Learning. https://ai4tl.blogspot.com/

[31] Draghi, M. (2024). The future of European competitiveness. European Commission. https://commission.europa.eu/topics/eu-competitiveness/draghi-report_en

[32] European Commission. (2025a). A competitiveness compass for the EU (COM(2025) 30 final).

[33] European Commission. (2025b). AI continent action plan (COM(2025) 165 final).

[34] European Policy Innovation Council. (2026). Draghi Observatory & implementation index [July 2026 update]. https://draghiobservatory.org/

[35] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). (2024). Official Journal of the European Union, L 2024/1689. http://data.europa.eu/eli/reg/2024/1689/oj

[36] European Commission. (2025c). Proposal for a regulation amending Regulations (EU) 2024/1689 and (EU) 2018/1139 (Digital Omnibus on AI) (COM(2025) 836 final). https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52025PC0836

[37] European Parliament. (2026). Digital omnibus on AI. Legislative Train Schedule. https://www.europarl.europa.eu/legislative-train/package-digital-package/file-digital-omnibus-on-ai

[38] Law of 23 September 2025, No. 132. Provisions and delegations to the Government on artificial intelligence. (2025). Gazzetta Ufficiale della Repubblica Italiana.

[39] Bradford, A. (2020). The Brussels effect: How the European Union rules the world. Oxford University Press.

[40] Schram, A. (2026, 20 September). Recursive, rapid, unregulated: Obama’s warning on AI. AI4TL – Artificial Intelligence for Teaching & Learning. https://ai4tl.blogspot.com/

[41] World Economic Forum. (2025). The future of jobs report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/

[42] IBM. (2024, 20 December; updated 10 July 2026). What is the AI skills gap? IBM Think. https://www.ibm.com/think/insights/ai-skills-gap

[43] Bradford, N. (2025). A deep dive into the World Economic Forum’s Future of Jobs Report 2025. SHRM. https://www.shrm.org/topics-tools/flagships/ai-hi/future-of-jobs-report-2025-deep-dive

[44] Goetz, G., Maddigan, M., & Schram, A. (2026, 8 June). Italy teaches business. It’s time to let students do business. Life is a Journey of Learning. https://albertschram.blogspot.com/

[45] University of Helsinki & MinnaLearn. (n.d.). Elements of AI [Free online course]. https://www.elementsofai.com