Compulsory School Teacher

In Valais classrooms, AI has already arrived: students use it for their homework, often without saying so and without any method. AI is not going to teach the class: it will transform everything around it, preparation, marking, homework itself, and make what actually happens in the presence of students more visible than ever.

The trade today

Compulsory schooling in Valais spans eleven years, eight of primary school and three of lower secondary (cycle d'orientation), and operates in the canton's two languages, with separate curricula depending on the linguistic region (PER on the French-speaking side, Lehrplan 21 on the German-speaking side). Teachers train at the HEP-VS (Valais University of Teacher Education), at its Saint-Maurice and Brig campuses. Between valley-floor school centers and village schools, some with multi-grade classrooms, working conditions vary enormously; the heart of the trade, though, stays the same everywhere.

A typical week breaks down into:

• Preparing teaching sequences: planning, choosing material, adapting it to the curriculum and to the real class
• Running the classroom: teaching, setting boundaries, motivating, managing group dynamics
• Differentiation: adapting the work for struggling students as much as for those moving quickly
• Marking and assessment: exercises, tests, essays, evaluations
• Individual follow-up: spotting disengagement, family difficulties, bullying situations
• Collaboration: special education, school leadership, speech therapy, school authorities
• Communication with families: parent meetings, interviews, messages, sometimes conflict

This fiche's badge puts automatable tasks at 30%, one of the lowest proportions in the series, alongside the mountain guide and the alpine cheesemaker. Unsurprising. Most of the trade plays out in the presence of children, and that part cannot be delegated to anything.

What AI is preparing

Preparing sequences speeds up. Generating a worksheet, a formative assessment, a text adapted to a curriculum theme, five variants of the same problem: this work, which eats up evenings and part of Sundays, will be largely assisted. The teacher keeps the scripting, meaning the essential part: deciding what the class should learn this week, in what order, through what detours. The tool produces material; it does not know where the class actually stands.

Differentiation becomes materially possible. Every teacher knows what should be done: three versions of the same worksheet, a simplified text for the non-native speaker, an extension for the student who finished in ten minutes. Almost no one has the time. Generative AI makes this kind of tailoring nearly instantaneous, and it is probably its deepest pedagogical contribution: it brings daily practice closer to what research has recommended for decades.

First-level marking. For closed-form exercises, automatic grading has existed for a long time. What is new concerns open-ended work: a first pass on the language, structure, and coherence of an essay can be generated, leaving the teacher the assessment that actually matters, the one that situates the text within the student's progress. The grade, the evaluation, and the placement decision remain professional acts that carry weight.

Students, for their part, did not wait for school. A growing share of homework submitted passes through a language model, with or without personal work behind it. Homework changes in nature: a text produced outside class no longer proves much about what a student can do alone. That unsettling observation is also an opportunity: it forces schools to rethink what they actually assess, and how.

Student data: the prerequisite

Valais public schools are subject to the LIPDA (RS/VS 170.2), and their students are minors: the combination demands the maximum caution found anywhere in this series.

No student data in consumer-grade tools. Names, grades, written work, family situations: none of it should pass through a service without a contract, without controlled hosting, and without a legal basis. A teacher who pastes a student's essay into a free tool is transmitting a minor's personal data to an unknown third party.

Tools validated by the canton. Compliance cannot rest on each individual teacher's vigilance: it requires a list of validated, contracted tools, hosted under a compliant framework, made available to schools.

Transparency toward families. What the school automates, with which data and which safeguards, must be explainable to parents in plain terms. Trust in schools is lost quickly and regained slowly.

What rises in importance for judgment

The pedagogical relationship. Authority, trust, the ability to get twenty-five teenagers working on a Friday afternoon: none of it comes out of a server. Education research keeps repeating it, the teacher effect ranks among the strongest determinants of learning. The more peripheral tasks get automated, the more visibly this relational core becomes the center of the trade.

Attention to weak signals. A student disengaging, a bullying dynamic taking hold, a family situation deteriorating: these signals show up in a real classroom, in hallways, at recess. No dashboard reliably surfaces them, and a dashboard that claimed to would raise surveillance questions a school should not sidestep.

Designing assessment. If the machine writes, what is left to assess, and how? More oral work, more in-class work, more assessment of process (drafts, steps, justifications): these shifts demand real pedagogical engineering, far subtler than chasing cheating.

Cultural and linguistic transmission. Chapter 13 of the essay stresses a point schools know instinctively: large models do not carry local heritage by default. They write standard French and standard German; Walliserdeutsch, spoken by some 80,000 people and nearly absent from written corpora, barely exists in them. The canton's history, its place names, its institutions (bourgeoisies, consortages, bisses), its bilingual culture: if the school does not deliberately pass them on, the tools students consult every day will drift away from them by simple statistical gravity.

Shaping students' judgment. Learning to check a claim, to distrust a text that reads too smoothly, to tell what one understands apart from what one merely copies: this critical literacy becomes a learning objective in its own right. Students will use these tools their whole lives. School is the only place where someone will teach them to do so with awareness.

Who keeps the final say?

Composite illustration. A lower-secondary teacher receives an unusually polished essay from an average student: flawless syntax, elevated vocabulary, none of the rough edges she knows from him. Rather than opening a case for cheating, she returns the papers ungraded and has everyone rewrite a paragraph by hand, in class, on the spot. She then takes the student aside for five minutes: what were you actually trying to say in that text? The conversation teaches her more than the paper did. The student leaves with an unexpected assignment: redo the homework using the tool, annotating what he kept, changed, and rejected, and why. (A fictional, composite situation, to be replaced by a real case during the incarnation pass.)

Job profile 2030

Three skills will need to enter job profiles explicitly; initial teacher training does not yet cover them.

The first is AI-assisted pedagogical scripting: steering generation tools to prepare, differentiate, and adapt material without abdicating the overall progression. Knowing what to ask for, evaluating what comes back, fixing what sounds right without being right. The pedagogical quality of generated material depends entirely on the eye that validates it.

The second is designing assessment in the age of language models: building tools that measure a student's thinking (oral work, in-class work, documented process) in a world where the finished product handed in from home no longer proves anything. This skill bears on the fairness of placement decisions; it cannot remain individual improvisation.

The third is critical AI education, paired with a strengthened role as a cultural go-between. Teaching the lucid use of these tools is one thing; embodying, in front of students, what the models do not carry, the local language, the local history, the habit of checking before believing, is another.

Territorial anchoring

School-age Valais has particularities no large model knows from the inside. First, cantonal bilingualism, with its linguistic border running through the canton and its bilingual programs; then Walliserdeutsch, the everyday language of tens of thousands of students, mostly spoken rather than written, which standard tools either ignore or mangle. Compulsory school is the institution that decides, generation after generation, whether this heritage gets passed on or erodes.

Small mountain schools, often multi-grade, are paradoxically well placed to benefit from AI-assisted differentiation: a teacher managing three grade levels in the same room needs exactly this kind of level-adapted material, every week, without spending nights on it. Properly equipped, the multi-grade village classroom becomes viable again; that is a direct argument for keeping local schools open in the valleys.

That leaves training. The HEP-VS, present in both linguistic regions, is the natural institution to carry this forward: building these skills into initial training, offering in-service continuing education, and documenting what actually works in real Valais classrooms rather than importing recipes.

What decision-makers must do now

For a school principal

Establish a written usage framework before the start of the 2026 school year, for teachers and students alike: what is allowed, what is forbidden, what must be disclosed. In parallel, take stock of practices already established in the school, because they exist, with or without a framework. A school that stays silent leaves every teacher alone facing cases of cheating, and every student facing implicit rules that shift from one classroom to the next.

For a cantonal official (Service de l'enseignement)

Publish a list of tools validated against the LIPDA, with framework contracts and compliant hosting, along with a cantonal doctrine on assessment in the age of language models. Both decisions exceed the reach of any single school; leaving them at the local level guarantees unequal treatment among students across the canton.

For the HEP-VS

Build AI-assisted scripting, assessment design, and critical literacy into initial training in both linguistic regions, and open an applied-research track on what Valais students actually do with these tools, at what age, in what language. The data is missing, and the canton's decisions over the coming years will only be as good as that groundwork.

---

Jérôme Deshaie is the founder of MCVA Consulting SA, an agency specializing in the AI transformation of organizations in Valais, and the author of Bisse Cognitif.