From doing to designing: how artificial intelligence is changing my work
Valentino Pietrobon · 20/11/2026
From a tool for single tasks to a collaborator for building materials, simulators and applications: an account of what I have done with AI, how I work with it, and what it could mean for schools.
Artificial intelligence did not enter my work overnight. Before this summer I was already using it, like many colleagues, for specific tasks: drafting the minutes of the teachers' board meetings from my notes, preparing group projects, designing the robotics and artificial intelligence workshops for my students and, even earlier, a training course funded by Italy's national recovery plan (PNRR) for the school's administrative and support staff. It was valuable help, but occasional: one question, one answer, one text to revise.
Besides teaching computer science from first to fifth year, for some time I have been the school's digital innovation coordinator, I draw up the school timetable, I manage around 1,500 Google Workspace accounts, and I run the school's AICA Test Center, organising and delivering courses and exams for digital skills certification. These are roles I love, but every day they take time and energy, often for tasks that repeat themselves unchanged. I was looking for a way to lighten their weight on everyday life, and I had a hunch: that AI could do much more than answer questions.
The two questions I started from
The first: can I renew my teaching materials? Course notes to update, exercises to rewrite, simulations I had wanted to build for years and never found the time for.
The second: can I get rid of repetitive work? Timetables, accounts, bookings: hours spent on tasks that come back the same every week and every year.
What changed in a few months
Between the end of May and the end of September, six new models came out, one after the other. The names are not the point: each step widened what I was able to do. In June some jobs stalled after a short while; by September the AI could carry the same project forward for hours without losing the thread, read entire documents in one go, use real tools, and check its own results before handing them over.
It is the difference between an assistant that makes suggestions and a collaborator that delivers finished work. What this means for people who write software I described in another reflection, "Zero lines of code"; here I am interested in what it has changed in my work as a teacher.
What I have built
Over these months I have revised or rewritten from scratch some twenty sets of course notes and built Sperimenta, a collection of around sixty interactive simulators for every year from first to fifth: pages that open on a phone or an interactive whiteboard and turn theory into something you can try with your own hands.
Alongside the materials, several tools were born, each from a concrete problem:
- Progetta, my environment for going from course notes to lessons and from lessons to tests, with competence-based assessment;
- Gestisci, for managing the life cycle of Google Workspace accounts for students and staff;
- Pianifica and Allena, still under way: one for the timetable, the other for simulating a class and training teachers to handle difficult situations.
Three tools, more than any others, have changed the working day, mine and not only mine.
Prenota was born for the school. It is an app, used from a phone or a computer, for booking labs, laptop carts and special rooms, with no more sheets pinned to doors or phone calls to find out whether a room is free. Every teacher can see at once what is taken and by whom; the app prevents double bookings and notifies whoever needs to know. The version I am finishing imports the school's real timetable, so classrooms also become bookable in the periods when they are empty, and bookings update themselves when the timetable changes. It is also the clearest example of what a school can now build to its own measure, instead of adapting to a product designed for someone else.
Jarvis is my personal assistant. Every night it reads the school circulars, the minutes of meetings, the emails of the inboxes I manage and the electronic register for me; in the morning, on my tablet, it gives me an overview of the day: lessons, meetings, parent-teacher appointments, deadlines. It sorts pending items into three lists: things to do, things to decide (applications, sign-ups, courses) and things to know. Before a class council it prepares a profile of the class; afterwards, it helps me tidy up my notes. The briefing is deliberately plain, almost like a weather report: the certain data are assembled without AI, and AI steps in only where it is really needed, to summarise a long document or flag something new. More than saving me hours, it takes away the feeling of constantly chasing information.
Jarvis also has a second role, designed for the school's open days. In this mode, without any personal data, it answers by voice the questions of parents and middle-school pupils: the weekly timetables of the different study tracks, the differences between the courses on offer, what we do in computer science, extracurricular activities, the options after the diploma. Projected on the screen, it also becomes a concrete way of showing what can be built with computer science. Questions about lived experience, such as what it is like to study here or how much studying it takes, it leaves to the students who are there.
Osserva, on the other hand, was born in the classroom. It is a tablet app that shows the students' photos: with two taps I record an episode while the lesson goes on. A remark that shows personal reworking, a correct answer, homework not done, a phone used during a test. At the end of the lesson the summary is ready to be copied into the register, and for each student there is a timeline from the start of the year, with positive signs on one side and negative ones on the other. It helps me assess on the basis of recorded observations rather than memories of the last week, and not to lose sight of the quieter students. A few colleagues are now trying it too.
Some of these tools are used every day, others are still growing. None of them, a few years ago, would have been within reach of a single teacher.
How I work with it, and where AI gets things wrong
What I have learned most about is method. The work always follows the same cycle: the AI proposes a plan, I decide what to change, cut or approve, the AI carries it out, then everything is checked, first with automatic tests and then by my own trial on the real case. The bigger the project, the more the rules written beforehand matter: every repeated mistake has become a rule that the AI rereads, so the work does not lose pieces from one session to the next.
The limits are real, and it is worth stating them clearly:
- it makes mistakes with confidence: a well-phrased error is still an error, and someone has to be able to read the result;
- it does not see hidden constraints: while working on the timetable file, a change that looked correct produced an error that only a test in the real program brought to light;
- it costs money: whatever can be done with fixed rules is done without AI;
- it needs protecting: credentials never stored, personal data only where needed, human review of security and access.
The teacher's expertise, in other words, does not become useless: it becomes quality control.
What changes for teaching
This, for me, is the most important part, and it applies to every subject.
Processes first, then tools. The real risk is cognitive offloading: the student gets the product without the learning. That is why the order matters: first secure the cognitive processes (understanding, reasoning, representing, evaluating), then develop competences to the level needed to judge what a tool produces, and finally introduce the tools, teaching them properly rather than ignoring them. For every activity the question is not "can it be done with AI?" but "which effort must remain the student's?".
Assessment becomes a conscious choice. The same test, marked in the same way, can produce different grades depending on the yardstick: a score per exercise, a blueprint table by topic and difficulty, a rubric with indicators, competence-based assessment. No yardstick is neutral, because each rewards different processes. Seeing them side by side, something nobody would do by hand, forced me to ask myself which one I use, and why.
What used to be too complex becomes manageable. It is not only a matter of speed. A series of simulators for a set of course notes, with their exercises built in, can be made in a day instead of months; several assessment yardsticks can be compared on the same piece of work. Choices we used to make by intuition become visible and comparable. AI does not decide for us: it puts us in a position to decide better.
Materials keep up with the present. Printed course notes age quickly, and in some subjects they age while you are writing them. With AI, updating them is no longer a job to put off until the summer. When, in recent weeks, recursive self-improvement (RSI) became a hot topic, meaning models that help improve themselves, within a few hours I updated the artificial intelligence notes for my fifth-year classes. I added a new section, the link to the closing reflections, and the part on learning from a verifier that is needed to understand it. Students study from a text that talks about the world they live in, not the world of a few years ago. Sources still have to be checked and the choices are still mine, but now there is time to do it.
The opportunities are for everyone. A simulator for your own topic, course notes updated with checked sources, tests and rubrics, exercises linked to competences, different versions of the same material for different needs, less time spent on paperwork. None of these uses requires knowing how to program. They require knowing exactly what you want and checking what comes back: the teacher remains the author of what goes into the classroom and the one who vouches for it.
An open question
There is one aspect that should not be left unsaid. Models, cloud services and platforms are in the hands of a few providers, almost all of them outside Europe, and prices and rules can change without warning. Between 12 June and 1 July this year, two of the most advanced models were suspended worldwide because of a United States export control measure: access to our tools can depend on decisions taken very far from us.
My own work also rests on external providers, and I do not hide it. For schools the question is what we hand over to others and with what guarantees: personal data only where needed, open formats, replaceable tools, simple rules where they are enough. It is a choice to be made with our eyes open.
In short
It would be nice to say that all these tools have freed up my afternoons. They have not: marking and school commitments still take up a great deal of time. Something has changed, though. Before, so as not to lose time I did not have, I would recycle and adapt the previous year's tests, which often ended up repetitive and poorly structured. Now I have time to really think them through: I prepare them with AI starting from the text of the course notes, paying attention to the variety of exercises, the balance of difficulty and the processes I want to assess. The same goes for lessons, which I can enrich with the simulators. The time AI gives back to me does not become free time: I devote it to what matters most, namely what my students should learn and how.
It is not yet what I imagined when I started building Progetta, but it is a good start. In the end, that is what the whole journey comes down to: AI does not take work away from teachers, it shifts it from doing to designing and checking.
The next step? Finishing the projects under way and turning Progetta into a study assistant that helps students with their homework.