More than half of teachers in Brazil, Chile and Colombia already use AI in their work, and not for just anything: they use it to prepare their lessons and very little to grade their students. It’s a sensible judgment. The problem is that each teacher applies it alone, because their institutions’ rules either don’t exist yet or nobody knows them.
Teachers use AI to prepare, not to grade
TALIS 2024, the OECD survey of lower secondary teachers, asks those who already use AI what they use it for. The pattern is the same in all three countries: generating lesson plans is among the main uses, and assessing or grading student work is among the least frequent.
View data
| Generate lesson plans or activities | Assess or grade student work | |
|---|---|---|
| Colombia | 80% | 39% |
| Brazil | 77% | 36% |
| Chile | 68% | 18% |
The gap suggests an implicit judgment. A lesson plan made with AI passes through the teacher’s hands before it reaches the classroom: the teacher reviews it, adjusts it and decides whether to use it. A grade, on the other hand, lands directly on the student. Teachers delegate where they keep the last word and hold on to what affects the student.
But the judgment belongs to each teacher, not to the institution
In Mexican higher education, the Ministry of Public Education’s ENIAG 2025 survey found that 60% of teachers use generative AI at least once a week. At the same time, 76.7% of teachers at public universities are not aware of any institutional rule on its use, and only between 25% and 51% have taken a course on it, depending on the type of institution. These figures come from a different level of education, but they point the same way: use has moved faster than the rules.
When the judgment is individual, quality depends on each teacher. What one learns doesn’t reach the others, and what another gets wrong goes unnoticed. The institution doesn’t need to invent the judgment. It needs to make it explicit and shared.
Three verbs to make the judgment explicit
Every teaching task falls into one of three zones: what can be automated, what AI can augment while the teacher decides, and what must be protected. Applied to real tasks:
| What to do | Why | |
|---|---|---|
| Generate variations of an exercise | Automate | Low risk and easy to check |
| Give feedback on closed-ended answers | Automate | The right answer does not depend on context |
| Generate the lesson plan | Augment | AI drafts it; only the teacher knows the group |
| Review performance data | Augment | AI organizes the data; the interpretation is the teacher’s |
| Grade work that has consequences | Protect | A mistake affects the student; the responsibility is the teacher’s |
| Support a student who is falling behind | Protect | The relationship is part of what is being taught |
For any other task, three questions are enough: what would a mistake cost? How well do you need to know the student to decide well? And is human interaction part of what you are trying to teach? The heavier the answers, the closer the task is to protect.
What academic leadership should decide now
- Make visible what is already happening. Ask teachers what they use AI for and classify those uses with the three verbs. It is the starting point for any policy, and it almost always shows that good judgment already exists in part of the team.
- Decide where the time saved goes. If nobody decides, more operational workload absorbs it. The value lies in reinvesting it in observing, guiding and giving feedback to each student.
- Measure learning, not just hours. Success is not how much time is saved, but whether the student’s experience changed and whether there are verifiable results.
The advantage won’t come from automating more than other schools. It will come from knowing what to automate, what to augment and what not to delegate.
