AI in education: the classroom of the future

Pretending students don't use AI is the worst answer a school can give. They use it. In the WhatsApp group, on the bike home, in the toilet during a test. So the question isn't whether AI is in your classroom, it's whether your teachers handle it better than a 14-year-old who has had ChatGPT as a homework buddy since primary school.

Meanwhile that same teacher is stuck on the real problem: not the technology, but the admin load. Lesson prep, differentiation, parent letters, report meetings, Magister, Cito. That's where AI buys time fast, faster than any "classroom of the future" demo.

Opportunities for innovation

The most interesting opportunity isn't an AI tutor that replaces the teacher. It's everything that currently keeps the teacher from teaching. A primary-school teacher spends a few hours a week on admin and correspondence that a proper AI process handles in ten minutes. A vmbo mentor preparing report meetings for 28 students wins back a full afternoon.

Same story at student level. Personalised learning has been a promise since the 1990s, but no one had time to build 30 practice routes for 30 students. With AI you make three variants in five minutes: a short one for those who get it, a longer one with more examples, a visual one with a diagram. Not a future scenario; a teacher in year 5 is doing it today.

The human factor remains crucial

The teacher isn't going anywhere. Not in 2026, not in 2030. A class of 28 needs someone who notices that Lieke is quieter today and that the group needs 20 minutes after gym to settle back at their desks. No model does that.

What AI does is hand the teacher's attention back. Less typing of lesson plans, more conversation at the desk. Not a romantic idea, a workload intervention.

AI in education: challenges and responsibilities

First challenge: digital literacy, and not only for students. A teacher who doesn't know how a language model invents what it doesn't know cannot explain why the physics answer was nonsense. Before you teach pupils anything about critical AI use, you train your own staff.

Second: plagiarism and assessment. You lose this fight if you bet on detection software, because it lags the latest model and produces false positives that hit the strongest students hardest. What works is redesigning assignments. Oral exams, presentations, in-class writing, process portfolios. Evidence not via "this essay was written by the student", but via "this student can think this thought out loud in front of me".

Third: outdated knowledge. Models hallucinate with confidence. For exact subjects, you don't put students on a chat model without a source check.

Fourth: privacy and data. Education works with children's data, special categories, social-emotional information. No consumer tool belongs in that pipeline.

The way forward

A school board that takes AI seriously does four things at once. One, train teachers, not with a one-off study day but with practical work sessions where they speed up their own lesson prep. The moment a year-3 teacher sees her parent letters finished in ten minutes is the moment AI is in.

Two, clear rules for student use, per subject and per level. No blanket ban and no blanket permission, because the gap between a vmbo-2 essay and a final HBO-year-4 thesis is huge.

Three, critical thinking higher in the curriculum. Source evaluation, model literacy, what is a hallucination, how do you verify a claim. Not a separate AI subject; woven into Dutch, history, biology, everything.

Four, a curriculum that can move. What you'll need in two years you don't know now. Build that flexibility in, otherwise you're at the same meeting table again in 2028.

Concrete AI applications for Dutch education

Four things that already work in a Dutch school or higher-education institution today. Report conversation preparation, where based on test results and journal notes, AI generates a first summary of a student's status, ready for review by the mentor. What used to take an hour now takes 10 minutes plus review time.

Parent communication, where generative AI writes standardised but personally-feeling parent letters (grades, school trip, sick-leave consultation, behavioural changes) in the school's tone of voice. Teacher reviews and adjusts. 80 percent time savings.

First-line feedback on draft essays, where AI gives students immediate feedback on draft versions of their work, so they can improve before final submission. Not as assessment (the teacher does that), but as a coach helping with the writing process.

Test analysis at class and level, where AI dashboards reveal patterns teachers otherwise miss: which topics score poorly across the whole class, which students get stuck on which sub-skills, which teaching approaches work for which group. And curriculum adaptation, where AI helps teaching teams design differentiating assignments and alternative exercises for students with different prior knowledge.

GDPR and AI Act in education

Educational institutions work with special GDPR categories: children's data, health data (extra support needs), and in some cases sensitive social-emotional information. Cloud AI with student data is only allowed under strict conditions: data processing agreement, EU-only storage, no training on student data. In 2026 this isn't optional, it's the floor.

The AI Act categorises some education AI applications as "high-risk", such as systems that determine admission or automate important assessments. For these applications: mandatory transparency, human supervision, documentation. Build AI Act compliance from day one, not patch in afterwards. Patching compliance in afterwards always costs double and produces a tool no one trusts anymore.

How a typical engagement unfolds in an education context

An engagement runs in phases. First a short analysis of which processes deliver the most time savings. Almost always that's parent communication or test analysis, not the shiny AI tutor the vendor wants to sell. Then a focused pilot on one process to test whether it fits your school culture, because a vmbo team operates differently from a HBO programme committee.

Only after that does scaling happen on what works, and stopping where the business case doesn't close. Where possible, plan key milestones around school holidays to minimise disruption. Honestly: if the pilot shows no time saved or no quality gain after three months, stop. Pushing on because budget is committed is how schools end up with two lost AI years.

Conclusion

AI in education is not about the classroom of the future, it's about the workload of today. Start with what keeps your teachers from teaching. Build a tooling stack around it that is compliant, trained on, and reviewed by humans. Assume students use AI and teach them to use it critically instead of banning it.

The schools that lead in three years won't be the ones with the most expensive AI licence. They'll be the ones where the year-4 teacher is finished with parent letters at four on a Friday, not at half past six.

Want to know which AI applications deliver the most value for your school or higher education institution? Take the free AI Scan and receive a personalised analysis in your inbox within 1 minute.