The AI Leap Foundation has published its First-Year Review, intended for everyone involved in the AI Leap program. It summarizes what was done and created during the school year, as well as the results achieved so far.
CEO Ivo Visak: Human Intelligence First, Artificial Intelligence Second
AI Leap’s pilot year demonstrated that the impact of AI depends on teachers’ skills and collaboration, students’ learning habits, and the agreements that schools are still developing, writes Ivo Visak, CEO of the AI Leap Foundation, writes AI Leap CEO Ivo Visak.
AI arrived in Estonian schools in late 2022, immediately after more capable language models became publicly available. During the first year of the AI Leap educational program, we began a nationwide search for a shared direction that would give teachers, students, and schools a clearer understanding of how AI should be used in teaching and learning. This school year, we have also made ITI (an application designed to support learning) available to students in grades 10–11. Unlike conventional AI chatbots, ITI is built around pedagogical principles and the safety of Estonian students.
AI Leap began in February 2025, when President Alar Karis announced the initiative. The Ministry of Education and Research, Estonian companies, and private donors joined forces around a shared understanding: The impact of AI on education is not a question for the distant future, but is already here, and we must begin addressing it immediately.
AI cannot be viewed as just another educational tool. It is a new socio-technical medium that is changing both learning practices and the questions which education itself must answer. How should learning be assessed when some traditional assignments no longer measure what we once believed they measured? How can AI make learning more visible rather than replace the process of acquiring knowledge?
The first year of AI Leap was a year of searching for answers to these questions. It also made one thing very clear: The best solutions emerge working alongside teachers and students, in real classrooms, through real successes and real failures.
The first year also showed that schools need to develop AI literacy and establish agreements about when using AI supports learning, when its use must be disclosed, when prohibiting it entirely is justified, and how a student’s own thinking should be assessed. Schools have very different starting points. These agreements will therefore not emerge from a single nationwide set of rules, but through collaboration among teachers, students, and school leaders.
In this sense, the most important outcome of AI Leap’s inaugural year was not a single methodology, training program or tool: It was the fact that teachers began thinking about these issues together. Teachers and upper secondary students helped us better understand where AI genuinely supports learning, where it creates only the illusion of simplicity, which existing educational practices need to be reconsidered, and which skills Estonian schools must consciously cultivate for the future.
The change is not limited to teachers using AI to create worksheets: Some schools have redesigned entire lessons so that students must demonstrate their reasoning and problem-solving process. Elsewhere, students have debated with AI, fact-checked its responses, and compared AI-generated explanations against primary sources. The focus has gradually shifted from asking how AI use can be detected to asking how students can be taught to use it critically.
Schools also need to know when actively prohibiting AI is in the best interests of learning. Here, too, simply establishing a rule is not enough: Students need to understand why, in a particular learning situation, working through the challenge will independently help them learn more than having a conversation with AI.
Usage data from the ITI student application has taught us one particularly valuable lesson. Approximately 12,500 students activated an account, and between 4,000 and 5,700 young people used the application each month this spring. ITI received an average satisfaction score of 3.07 out of 5. These figures reflect a much broader challenge, which is that an AI application designed to support learning must compete with online tools whose purpose is to give users an answer as quickly as possible. ITI is designed to do the opposite: Not to do the work for students, but to encourage them to think for themselves. This is precisely why the most common criticism from students was that ITI asked too many follow-up questions instead of simply providing an answer.
One of the most important lessons from the first year is therefore that an AI application grounded in research-based principles of learning must also be sufficiently clear, flexible, and user-friendly. Otherwise, students will inevitably choose an easier alternative, even if it is less beneficial to their learning.
At the same time, the usage data does not support the simplistic assumption that young people only use AI to obtain ready-made answers: Learning-oriented and mixed-use patterns predominate among active users. Feedback also showed that students expect very different things from ITI: one student wants guidance, another wants a quick explanation, and a third wants to check an answer. A single rigid conversational approach cannot meet the needs of every learner or every situation.
The long-term effects of learning with AI on students’ motivation, perseverance, self-regulation, and learning strategies are being examined in a study led by the University of Tartu which involves more than 9,000 students. First-year usage data reveals patterns of behavior and user experience, but it does not yet provide a basis for claiming that ITI has improved learning outcomes. The first substantive findings are expected from the research team (headed up by neuroscientist Jaan Aru) later this year.
Alongside the ITI learning application, students also have access to free language models, and in several subjects, pedagogical approaches developed over decades no longer work as intended. Educational psychologists Grete Arro and Kati Aus from Tallinn University, Anneli Veispak from the Estonian Business School, and Gerli Silm from the University of Tartu have also made important contributions to understanding this shift.
During the 2026/27 school year, AI Leap will expand to include teachers in grades 5–9, as well as vocational education students and teachers, because the impact of AI extends along the entire educational pathway. We will need to adapt our support to the different needs of lower secondary schools, upper secondary schools, and vocational education. We also aim to make ITI more flexible and more responsive to individual subjects and learners.
None of this means that we already have all the answers. Quite the opposite: We have begun asking the right questions at the right time, and many discussions that other countries have yet to begin having are already underway in Estonia. Perhaps the most obvious lesson from AI Leap’s pilot year is also the most important: A new technology can be adopted within a few months, but changing teaching and learning practices takes years.
Access to AI alone does not help anyone learn more effectively, because learning still takes place in the learner’s own mind. The teacher’s role is to create learning situations in which AI helps to establish the conditions for meaningful learning. It is also the teacher who decides when actively setting AI aside is in the best interests of learning.
The AI Leap First Annual Review (in Estonian, English version coming soon) provides a more detailed account of what was done and created during the school year, the results achieved, and where the program is headed next.