AI as a catalyst for a new learning culture

Understanding and using AI

According to a study commissioned by the health insurance provider Barmer and conducted by the Sinus Institute in 2025/2026, 19 percent of young adults feel threatened by AI.[1] The teaching staff at Stadthagen Technical School agreed early on that they would counter this uncertainty with specific employability skills. Those who understand AI and are able to evaluate results obtained via AI will perceive they are empowered by the technology rather than at its mercy. With this in mind, Stadthagen Technical School – part of the Ludwig Fresenius group of vocational schools (LFS) – set about its task.

The school specialises in mechanical, electrical and construction engineering and has an annual cohort of 60 learners. This means that it operates in a specific environment within higher vocational education and training. The target group differs significantly from learners in initial training. The school’s trainee state-certified technicians have already completed a VET qualification and gained relevant practical experience upon arrival. Although they are intrinsically highly motivated, they are often sceptical of higher education – particularly if a direct benefit for professional practice in a role such as site manager or design engineer is not immediately obvious. For this reason, AI must not be integrated into teaching as an abstract additional subject. It needs to act as a tool which informs everyday professional work (for example in the areas of construction, structural engineering and project management).

Use of AI requires an altered understanding of learning

AI is already enshrined within the curriculum at Stadthagen Technical School. Such a rapid implementation was due to the fortunate structural circumstance that the school had already started radically modernising its curricula prior to the AI wave. The conventional range of subjects was dissolved in favour of cross-cutting learning situations. Learning now occurs in action-oriented scenarios instead of via the isolated teaching of “production engineering”, “manufacturing and design” or “information technology”. One typical example of this is the planning of a ventilation system for a residential building. The learning scenario here links mathematical calculations, technical standards and documentation in a process-oriented manner. Only core subjects such as mathematics or German remain in place as cornerstones.

It was precisely this structure which proved to be the perfect prerequisite for AI integration. This is because AI simply acts as another powerful resource in the problem-solving toolbox in such practice-related learning situations. If a task involves developing a complex technical concept, learners are given the option of using AI for the purposes of structuring or brainstorming. In some cases, it is even mandated.

Deployment scenarios for AI-based learning

The introduction of AI to the classroom risks triggering a phenomenon which the Deutsche Telekom Foundation’s AI in Education Trend Monitor (2025) explicitly warns against: deskilling. Professional expertise will wither away if learners delegate complex calculations or design specifications to an AI system via the click of a button without fully grasping the pathway to the solution.

The didactic response to this at Stadthagen Technical School is “human in the loop”, i.e. a person checks AI-generated outcomes and intervenes to correct process sequences if necessary. The aim is to train learners to become AI experts, not AI operators. Ludwig Fresenius Schools use their own tool for the technical implementation of this in compliance with data protection regulations. This tool is known as “Fresi” (Next Generation AI). Fresi is more than a mere interface to common language models. It uses RAG (Retrieval Augmented Generation) technology. And this is crucial for teaching. The language competency of the AI is linked to a verified knowledge database containing the school’s own scripts, standards sheets and technical documentation. AI is not able to fabricate anything because RAG forces it to formulate responses on the basis of the teaching materials which have been made available. This reduces hallucinations, albeit without eliminating them entirely, and this is precisely where the added pedagogical value kicks in. In a technical environment in which standards (in Germany DIN, VDI) and safety regulations are absolute, the sycophancy of large language models poses a risk. This can be illustrated by a vivid example from the engineering module, the “glove paradox”. If learners ask generative AI about health and safety measures for a lathe, the system frequently proposes gloves as a stock response on the basis of the general safety data found on the Internet. But trainee technicians need to know that gloves and rotating parts are an extremely dangerous combination (risk of entanglement). Such errors are deliberately provoked in lessons. Learners take on the role of testers, attempting to trick the AI into making mistakes or seeking critical security vulnerabilities in the output. This moment underlines the necessity of human expertise. Only those with specialist knowledge can recognise a hallucination which may sometimes be potentially deadly. This bolsters self-confidence enormously.

Alongside its deployment for technical assessment and examination, Fresi can also be used for new learning pathways which foster the critical thinking skills encouraged by the Conference of the Ministers of Education and Cultural Affairs (KMK). Via appropriate system prompts, Fresi operates like a Socratic dialogue (flipped testing), i.e. acting as a tutor rather than as a solution machine. The learners enter a problem, and AI poses counter questions to guide them towards the solution without giving it away entirely.

Transmedial learning, whereby complex technical content is converted into other formats by AI, is equally important. Learners transform a dry PDF text on building law into a script for a podcast, flashcards or a mind map. Learners must have gained a thorough understanding of the material in order to assess whether the AI-generated podcast captures the core of the specialist subject matter.

The KMK’s call for co-activity between humans and AI has been embraced within the daily routine at Stadthagen Technical School. This is exemplified by the following scenario. In BIM-based modelling of a residential building, AI generates the textual building specifications on the basis of the technical benchmark data. The didactic benefit is clear. Because AI does the paperwork for the learners, they have more time to focus on the more technically challenging aspects of 3D modelling and on clash detection within the model. AI does not lead to laziness. It actually creates scope for carrying out cognitively demanding engineering tasks. Humans plan and check, and AI documents.

Evaluation of AI-generated results

This integration of AI forces a new type of performance assessment. If a final result may theoretically originate from an AI system, it loses value as the sole criterion for assessment. Evaluation of the learning process then becomes more important. Instead of attaching confusing chat histories as attachments, learners are challenged to write a reflective report on their use of AI. They need to set out the prompt strategy they have used and must explain where they have identified hallucinations and how they have professionally adapted the output. Someone whose report states: “AI proposed the DIN 4102 fire safety standard, but I corrected that to the more current DIN EN 13501 standard” will receive a better mark than those who submit a perfect text upon which they have not reflected. Critical AI literacy is the object of assessment here.

Working towards greater real learner autonomy in the classroom

The introduction of generative AI at Stadthagen Technical School represented a cultural shift rather than merely a technical update. Teaching staff have abandoned the notion that school is simply a repository of knowledge and instead increasingly view themselves as trainers of competency managers. The most important future competency is the ability to professionally evaluate the result of AI systems, not writing prompts. AI does not lead to deskilling if it is incorporated in a way which makes didactic sense. It frees learners from routine activities and creates time for more complex tasks. The conclusion to be drawn from practice is clear: AI is outstanding at replacing mediocrity. Those who only provide standard responses will become obsolete. Our education and training remit is therefore to turn learners into excellent qualified skilled workers who master the machine as a tool and are not replaced by it. The journey to this point proceeds by actively exercising critical autonomy in the classroom, not with AI bans.

Boris Singer
Deputy Headteacher, Technical College of the Ludwig Fresenius Schools, Stadthagen

Translation from the German original (published in BWP 2/2026): Martin Kelsey, GlobalSprachTeam, Berlin