Building AI capability in the team

Capability Is Not a Training

The first reflex when AI is coming into the house is almost always a training. Half a day, a few slides, a prompt cheat sheet, done. The problem is not that training harms. The problem is that on its own it does almost nothing. Two weeks later most of it is forgotten, because nothing sat between the seminar and the actual work for the learning to attach to.

Capability forms where someone solves a real task with AI, gets a poor result, understands why, and does it better next time. That loop cannot be shortcut in a seminar. It needs real cases from your own day-to-day, repetition and someone who catches questions as they come up.

AI capability is not a training you tick off. It is a habit that grows on real tasks and is maintained in operation.

So the right question is not "which training do we book" but "on which real tasks does the team practise, and who supports that". A course can give the entry. Capability comes afterwards.

The Stages From Curiosity to Practice

A team does not become capable overnight. It moves through stages, and each one feels different. It starts with curiosity: people try things, marvel at good answers, get annoyed at bad ones, but none of it is routine. Then comes reliable use, when AI becomes an obvious tool for standard cases and is no longer a small expedition each time. The most important stage is critical judgement: only when people recognise when an answer sounds plausible but is wrong does the use become safe. At the top sits applied practice, where a team improves its own cases and rewrites its own rules.

The stages build on each other. Skip critical judgement and you get a team that trusts AI where it should not. That is exactly where the errors arise that get expensive later.

The goal is not for everyone to reach the top stage. The goal is that nobody stays stuck at curiosity while already working productively with AI.

Who Needs Which Depth

A common mistake is to have everyone enter at the same depth. That is expensive and rarely on target. Users in the business need to use AI safely and interpret results, no more. A business owner has to judge quality and exceptions, deciding what a tolerable error is. Multipliers show others the way and catch questions before they turn into a block. And leadership has to judge value and limits realistically, to set sensible rules rather than acting on gut feeling.

This distribution decides whether capability building stays affordable. If everyone is supposed to do everything, nobody gets genuinely good. If each role practises what fits, the whole thing meshes.

The EU literacy obligation in Article 4 of the AI Act requires a sufficient level of AI competence among the people who work with the systems anyway. Building capability by role meets that requirement as a by-product, not as a tiresome duty.

How to Build Capability That Lasts

Before you book a training, align the build with these five points. They cost more attention than budget, and that is exactly why they work.

• Real tasks, not demo examples: Practise on cases from your own day-to-day, not on clean examples that always work.
• Set roles and depth: Clarify who only needs to use, who needs to judge and who needs to show, and practise accordingly.
• Clear usage rules: Record what may go in, what may not, and when a human checks the result before it takes effect.
• Name multipliers: Appoint people who catch questions so no uncertainty in the team fizzles out.
• Repeat the practice: Plan recurring, short practice rounds instead of relying on a single event.

Core sentence: capability building is in small part a training project and in large part a question of practice, roles and rules. If you only train, you pay the rest later in errors.