AI training for employees: a four-level plan (Charlemagne would approve)

You bought the licenses, people have access, but a month later only three of them use AI, each in a different way. It almost always happens, and the reason is simple: nobody taught them how. AI training for employees isn't a motivational talk: it's the step that decides whether the investment pays off or stays on paper. Here's what the law asks, a four-level plan, how much time it takes and how to fund it.

Charlemagne, schools and badly copied books

In 789 Charlemagne issued a long capitulary known as the Admonitio generalis, the "general admonition" to the bishops, clergy and officials of his realm. Among its many provisions is one that historians of education hold dear: monasteries and cathedral churches were to open schools where boys would learn to read, along with psalms, chant, reckoning and grammar (see the overview of the Admonitio generalis).

The reason was very practical. In a world where books were copied by hand, every careless scribe introduced errors, and those errors multiplied copy after copy. So the capitulary asks for books to be properly corrected and for copying to be entrusted to mature, skilled people: a faulty text produced faulty prayers, or, in today's language, wrong results that look right. Charlemagne's answer was neither to ban books nor to rely on goodwill. It was a literacy program, together with systematic care for the quality of texts.

You may say: Professor, what do Carolingian schools have to do with my purchasing department? Quite a lot. Today every employee has access to a tool that produces fluent, tidy, convincing text. And sometimes wrong text. Whoever can't spot the error copies it into the customer email, the quote, the report, and the error multiplies. Just like in the manuscripts.

Well, your company today needs its own little Carolingian school. And this isn't just a historian's obsession: Europe asks for it too, using the same word. Literacy.

What the law asks: Article 4 of the AI Act

The AI Act devotes Article 4 to AI literacy. In its original version, applicable from February 2, 2025, it asked providers and deployers of AI systems to ensure "to their best extent" a sufficient level of skills among their staff.

Regulation (EU) 2026/1744, the so-called Digital Omnibus, in force since July 27, 2026, rewrote the article: today those who use AI systems must take measures to support the development of AI literacy among their staff and others using the systems on their behalf, taking into account skills, experience, education and context of use. The new text states that it does not require guaranteeing a specific level for any individual.

In practice: the obligation remains, but it's an obligation of means, not of result. You must be able to show you did something sensible and proportionate. A usage policy, role-based training and a record of who attended are the simplest way to do it.

In Italy, Law 132/2025, Article 11, adds that AI at work must be safe, reliable and transparent, and that employers must inform workers in the cases provided. Informing and training aren't the same thing, but they go together.

The four-level plan

Not everyone needs to know the same things. Here's a structure that works in small businesses.

Lengths are indicative and should be adapted. The principle, though, is fixed: little, often and on real cases.

Level 1: the alphabet for everyone

The basic module should leave everyone with four certainties:

1. generative AI predicts plausible words; it doesn't "know" things: it can invent facts, figures and sources;
2. every output must be checked before it leaves the company;
3. some data never goes in (the traffic light from our ChatGPT at work policy);
4. whoever uses the tool remains responsible for what it produces.

Level 2: the team's own cases

This is where the difference is made. A generic course shows how to write a poem about Mondays; a useful course shows accounting how to check a reconciliation, sales how to draft a quote, the warehouse how to query internal procedures. Bring real documents (stripped of personal data) and have people work on their everyday tasks.

Level 3: internal champions

Every area needs someone to ask "is this OK?". Champions collect the best use cases, flag problems and keep instructions up to date. They link the people who do the work and the people who decide.

Level 4: leadership

Those who decide on investment must understand costs, risks and rules: otherwise they buy licenses nobody uses or block everything out of fear. Useful reading for this level: our complete guide to AI for small businesses and what the AI Act means for SMEs.

How to measure whether training works

Attendance isn't enough. Measure instead:

• real use: how many people use approved tools each week, a month after the course;
• quality: a small practical test, such as spotting the error in an AI answer;
• time: do the tasks chosen as use cases take less time?
• questions: how many reach the champions (lots at first is a good sign).

And keep a simple record: date, topic, participants, materials. It's evidence of the "measures" Article 4 of the AI Act asks you to take.

How to fund training

In Italy the main route for continuing training is the joint inter-professional funds, which a company joins by directing part of the contributions it already pays. Several funds have opened AI-specific lines. Two verified examples from Fondimpresa, the largest fund:

• Call 4/2025 "Artificial Intelligence", with 5 million euros, closed early at the end of February 2026 because demand exceeded funds: a sign that interest is high and you need to move early;
• Call 4/2026, published June 30, 2026 with 30 million euros, funds advanced-skills plans including an "Artificial intelligence and virtual reality" area, with priority for employees of micro and small firms; plans are submitted by qualified training bodies from December 1 to 22, 2026.

Other funds (for services, cooperatives, professional firms) have also published AI calls: check which one your company belongs to and ask your payroll adviser. The full picture is in our guide to AI incentives for small businesses.

Mistakes to avoid

• One webinar for everyone. Forgotten within a week.
• Training before choosing the tools. People learn on a tool they won't end up using.
• Teaching only prompt tricks. Without the part on checking and data, you train fast, careless users.
• Ignoring fear. Many people fear being replaced. Saying so openly, and explaining how their work will change, makes training far more effective. More in AI and work: which tasks change.

Typical scenario: a 40-member services cooperative

Picture an administrative-services cooperative with 40 people. Management spends half a day on level 4, then everyone attends a two-hour basic module. Each office holds three two-hour sessions on its own cases: payroll on client communications, front office on standard replies, accounting on reading documents. Two champions follow a longer track. Everything is funded through the cooperative's inter-professional fund. It's a typical scenario, but it shows that a serious plan takes a few weeks, not a year.

Let's talk

At Penzai we design tailored AI training built on each team's real cases, for small and mid-sized businesses and professional firms. See our AI training service or book a free intro call: we start from what your people actually do every day.