AI Training: Paths to AI Knowledge

AI training in one sentence

AI training covers all the formats employees use to apply AI systems in a professionally sound, legally compliant, and critical way – from multi-month certified full-time programs to IHK courses and university certificates, all the way to structured self-paced learning paths in everyday work.

What's new about this above all is one thing: training on AI is no longer just a career question. Since February 2025, the EU AI Act has required companies to ensure that employees who use AI systems have sufficient AI competence at their disposal. What used to be a nice-to-have has become a task you must document. The details are covered in Training Obligation under Art. 4 EU AI Act.

The Four Competency Levels

Before choosing a format, it's worth asking: which level is actually needed? Most misguided investments happen because a team books a data science course when what it actually needs is application skills.

For an average company, the biggest lever lies in the two middle levels. The entry point into basic understanding starts with What Is Artificial Intelligence?; application skills build up fastest via Writing Better Prompts .

Formats Compared

For the broad workforce, the last row is usually the most cost-effective: a curated path made up of learning units that build on each other, with a knowledge check at the end. That's exactly how the knowledge hub here at buzzmatic.com/wissen is built – sorted by topic area and difficulty level, with a quiz and completion certificate for each level. For the legally required basic understanding and application skills, this is often enough; for design and specialist skills, you add a deeper format as needed.

How to Spot Serious Offerings

The market is crowded, and quality varies enormously. Five check questions reliably sort the wheat from the chaff:

1. Is it practiced or just shown? Without hands-on tasks using your own material, nothing sticks. Demonstrations are entertaining but ineffective.
2. Are limitations addressed? An offering that doesn't cover hallucinations, data protection, and error rates is selling enthusiasm instead of competence. Why that matters is shown by AI Hallucinations.
3. How old is the material? Screenshots and model names give away the age of the content. In a field that changes every six months, two-year-old material is worthless.
4. Who is teaching it? Someone who uses AI in their own work – or someone who only ever talks about AI?
5. Is it tool-neutral enough? A course that covers only a single product is product training. That can be useful, but it doesn't replace training.

Funding: What's Generally Possible

Germany funds training through several channels. The details change, but the structure stays stable:

Education voucher (Bildungsgutschein). Issued by the Federal Employment Agency or the Jobcenter, mainly for jobseekers and people at risk of unemployment. Requirement: the provider and the program must be accredited under the Accreditation and Approval Regulation for Employment Promotion. Without this accreditation, there's no funding, no matter how good the course is.

Funding for employees. Even people in ongoing employment can be funded if the qualification responds to structural change – AI is a textbook case for this. Depending on company size and circumstances, course fees and in some cases wages are subsidized. The Employer Service of the Federal Employment Agency is the point of contact.

State programs. Almost every German state has its own training vouchers or education bonuses with different subsidy rates. It's almost always worth checking the website of the relevant state development bank.

Educational leave (Bildungsurlaub). In most German states, employees are entitled to paid leave for recognized training, typically a few days per year. Whether a specific program qualifies is stated in the course description.

Important: funding almost always has to be applied for before the course starts. Approval after the fact is the exception.

Skill Matrix: Making Team Needs Visible

If you assign training by word of mouth, you end up training the wrong people. A skill matrix makes needs visible within an hour and then becomes the basis for budget and documentation.

Here's how to proceed:

1. List roles instead of people. Editorial, sales, accounting, customer service, development – five to ten rows are enough.
2. Use four to six competencies as columns. For example: basic understanding, prompting, data rules, result review, automation, tool configuration.
3. Assess the current status. Three levels: doesn't know it / uses it occasionally / works confidently with it. Self-assessment plus a cross-check by the manager is enough; elaborate tests bring little added value.
4. Set the target status per role. Not every role needs the highest level everywhere – that's the actual value of the matrix.
5. Prioritize the gaps. Close what's legally required first (basic understanding, data rules), then whatever brings the biggest time savings.
6. Update every six months and document the results – this doubles as part of your competency evidence.

The matrix needs to be closely tied to your usage rules in content terms: whatever you define in the AI Policy for Companies in terms of data classes and review obligations must actually show up in the training.

What a Certificate Is Worth

An honest assessment: a certificate proves attendance and – for tested formats – passed knowledge. It doesn't prove that someone actually uses AI productively in their own work. For the job market, established issuers like the IHK and universities count for more than platform badges.

Still, even smaller credentials serve a purpose: they are documentation. For the EU AI Act's competency requirement, you need to be able to prove that training took place – attendance certificates, quiz results, and a maintained skill matrix satisfy that. The value here lies less in prestige than in being able to prove it.

Conclusion

AI training works when the question of the required competency level is answered first, and only then the question of format. For the broad workforce, curated learning paths with a knowledge check are usually the most cost-effective route; for individual roles, IHK or university formats pay off; and for teams with shared tasks, in-house training with follow-up support works best. Funding is available, but it must be applied for in advance. And since the EU AI Act: training is no longer optional, but something you must be able to document.