The initiative
Ask what AI can bring to education, rather than dismiss it.
In schools, AI first arrived as a disruption: students cheat, homework becomes suspect, models hallucinate. This initiative turns the question around. AI will not replace teachers. It can strengthen their practice, so that more students succeed.
Teaching under pressure
A teacher is expected to build their own teaching. Many situations leave no time to do so.
- New teachers
- They start with a class before they have built a single progression.
- Contract teachers
- Hired during the year, they often take up the post the same week.
- Replacements
- They must adapt to a class, a level and a progression already under way.
- Return to work
- The same adaptation, after months away from the classroom.
- New class or level
- In the same school or after a transfer, the work starts again.
These situations share one feature: the teacher must quickly produce a progression they had no time to build, while already teaching. Students with specific learning disorders add to the difficulty, and so do teaching conditions, which often depend on the school’s socio-economic environment.
Curricula that keep changing
Each revision can undo the long work of building a progression.
- Former content becomes obsolete.
- Former practices persist, and resist change.
- The orientations of the pedagogical council evolve.
An example from France: the new lower-secondary mathematics curriculum (cycle 4), published on 5 March 2026, comes into force in the first year of the cycle in September 2026, in the second in 2027 and in the third in 2028. A lower-secondary mathematics teacher has to rebuild their progressions three years in a row.
Learning paths that break
Some students reach a class without the prerequisites it assumes.
- Truncated paths: some concepts were never taught, because of a missing teacher, a strike, a lockdown or a period out of school.
- Challenging classes.
- Students in difficulty.
Achievement hidden by marking scales
Marks depend on each class’s own scale, so one school cannot be compared with another.
The aim is not to rank schools. It is to place what students have learnt against the expectations of the curriculum, competency by competency, rather than on marking scales specific to each class.
Many supports, disconnected
No teacher starts from a blank page. The supports exist, and they are often good. They are not linked together.
- Curricula
- The official curriculum sets knowledge, competencies and end-of-year expectations. It says what to teach, not how to organise it over time. When it changes, nothing links the old version to the new one.
- Training
- Initial training comes before the post, continuing training from time to time. Neither gives a concrete progression for the class in front of the teacher. A contract teacher hired mid-year has often had neither.
- Expertise
- Practitioners know which sequences work and which errors students make, but this knowledge travels poorly. Didactics research has described many typical errors; its findings rarely reach the classroom in a usable form.
- Supervision
- The pedagogical council and the inspectorate guide practice, with limited means: one inspector follows several hundred teachers, with no hold on the tools teachers use every day.
- Resources
- Official banks, textbooks and publishers offer abundant content. Each organises it in its own way, with its own coding of the curriculum.
An ecosystem to connect
Taken together, these supports are considerable. What is missing is not one more resource, but alink: a common curriculum reference, in a form tools can use, to which every resource, item, teaching sequence and typical error can be attached. With such a link:
- a curriculum revision spreads to the progressions that depend on it;
- practitioners’ expertise is shared between colleagues in a reusable form;
- research findings feed directly into error diagnosis;
- supervisors can validate and share reference progressions;
- existing resources are integrated without being rewritten.
The tool we propose replaces none of these supports: not the curriculum, not training, not the inspectorate, not textbooks. It links them, and puts the result in the teacher’s hands at the moment they build their teaching.
Why it is possible now
This linking role was long too costly to build. Three changes make it realistic.
Agentic capabilities
AI can chain differentiated actions into one procedure: from curriculum to progression, to assessment plan, to items, to marking and learning analysis.
A sharp fall in development costs
A tool linking curriculum, progression and assessment, once reserved for large institutional projects, is now within reach of a small team.
Sovereignty and cost control
Open-source models, hosted locally, keep data and processing under control, without depending on a provider’s pricing. Models are chosen on measured criteria.