AI and skills: what Norway teaches us about AI Governance

Whenever a technology reduces the time needed to do something, the advantage is easy to measure. How much time we have saved. How many documents we have produced. How many tasks we have automated. How many people use the new tool. Far harder to measure is what, in the meantime, we have stopped exercising. Nobody thinks about this. Or rather, almost nobody. Because (news from the past few days) Norway has just decided to tackle this problem, starting with its schools. From the new school year, pupils in years one to seven will largely be shielded from direct use of artificial intelligence in their schoolwork. From years eight to ten, access may increase gradually, under teachers’ guidance. In upper secondary school, students will instead have to learn to use these tools critically and appropriately, also in preparation for university and work (Utdanningsdirektoratet). It is an interesting progression because it introduces a criterion missing from much of the debate on artificial intelligence:

Competence first, then delegation.

And I believe this question now concerns far more than schools.

Results can improve while competence declines

A student has to write an argumentative essay. Before even getting to the words, they must understand a question, gather information, take a position, order the arguments, spot the contradictions, decide what to cut and rewrite. In the end there is a text, but much of the learning happened along the path needed to produce it. Artificial intelligence can compress that path enormously. The OECD Digital Education Outlook 2026 brings together studies in which the use of general-purpose GenAI tools improves the quality of students’ work during the task. When they later have to sit a test without assistance, that advantage disappears and in some cases reverses. The OECD also links the phenomenon to the risk of metacognitive laziness: a reduction in cognitive effort that can affect skill acquisition (OECD Digital Education Outlook 2026). The interesting finding concerns the possible gap between performance and capability. We can achieve a better result without having become more capable of producing it. For schools it is clearly an educational problem. For a company it should be, first and foremost, a strategic one.

What schools are seeing before companies

Imagine a young analyst using AI to summarise a market before having learnt to read data critically. A copywriter who entrusts a company’s voice to a model before learning to recognise what makes that voice consistent. A manager who uses an assistant to compare scenarios and build recommendations without knowing well enough the assumptions on which those recommendations rest. The evidence gathered in schools does not automatically prove that the same happens in companies. It does, however, point to an organisational risk worth taking seriously: Granted, all this is far harder to see than an obvious error. But an error can be corrected. A capability that the organisation gradually stops exercising, on the other hand, may only become visible when it is needed again.

The first decision concerns who stays in control

In January, at BETT in London, Education Minister Kari Nessa Nordtun stressed the need to be far more intentional about when to introduce different technologies, drawing a sharp distinction between the needs and capabilities of a six-year-old and those of a sixteen-year-old (Speech at BETT 2026). Behind this position lies a problem many organisations already know. Technology makes its way into infrastructure very quickly. That is partly why adoption can happen before anyone has really decided how it should happen. This is the ground on which Shadow AI grows, and it is one of the reasons why a corporate AI Policy must make explicit the authorised tools, usable data, oversight and accountability. Policy only reaches part of the problem. Before that, we need to establish what we want the technology to do within the organisation.

The question that should reach the board

When we talk about AI Governance, we tend to think of policies, system registers, security, compliance, accountability and output control. All necessary, of course. Upstream, however, sits an even more strategic decision: what cognitive capital do we want to keep inside the company? Every organisation holds knowledge that must continue to be widely shared. Skills that someone must be able to exercise even without an assistant, and that no one can afford to lose. That is why an AI adoption plan should also ask:

- Which skills must a person have before using the tool?

- Which activities do we want to accelerate?

- Which capabilities do we want to keep training?

- Where must human verification remain?

- Who can stop its use when the tool produces results that appear efficient yet are organisationally harmful?

All questions of governance, because they concern the distribution of decision-making power.

The time to choose has come

Schools have a particular characteristic. Their end product coincides, at least in part, with the person’s capability. That is why, for schools, the urgency is especially clear and evident. In companies, however, this phenomenon may be hard to observe. If a presentation is good, it gets used. If a report arrives faster, the process looks efficient. If a text works, it gets published. The loss of competence stays hidden until the moment we need precisely that competence to realise the machine is getting it wrong. And this is where Norway offers an interesting lesson. It is trying to establish the sequence of automation. First, some fundamental skills. Then, knowledge of the tool. Next, increasingly autonomous use. Finally, the ability to use it within complex contexts. A simple sequence, yet one that few are applying.

Every automation should have a skill to protect

Over the past two years, we have measured the adoption of artificial intelligence mainly in terms of speed. Hours saved. Tasks automated. Processes accelerated. Cost per output. Useful metrics. But Norway suggests another:

Which capability do we want to protect while we automate this process?

It is a question a CEO should choose to ask before approving any significant adoption of AI. Because every automation produces a visible gain. Some also change what people are still able to do. Mature governance should be aware of both. The first map of AI within a company, then, could start from skills rather than software, asking which ones we can delegate, which we can amplify, and which belong to a person’s responsibility. Norway is trying to answer this question to safeguard children and teenagers. Many companies will soon have to start asking it too, to safeguard their own operations. Because when a machine can do something in our place, the real decision concerns what we want to remain capable of doing without it. Today, tomorrow and the day after.

Domande frequenti

Has Norway banned artificial intelligence in schools?

No. The new recommendations follow an age-based progression. From years one to seven, direct access to GenAI should be avoided in most cases. From years eight to ten, it can be introduced gradually under teachers’ guidance. In upper secondary school, students should learn to use it critically and appropriately. (Utdanningsdirektoratet).

Does artificial intelligence worsen learning?

It depends on how it is used. The OECD highlights that general-purpose tools can improve performance during a task without producing an equivalent improvement in learning. Uses designed with a precise pedagogical purpose show more promising results. (OECD Digital Education Outlook 2026).

What can the Norwegian model teach companies?

A principle of sequence. Before automating an activity, it is worth establishing which skills the organisation wants to retain, what role to assign to AI, which checks to keep in place and who retains responsibility for the decision. These are some of the central questions of an AI Governance system.

Fonti e riferimenti

Scritto il 2 October 2026

Before we delegate to AI, we must decide what we want to remain capable of: the lesson from Norway