Estonian students have rendered a verdict on the state's national AI tutor, and it is not flattering: "a lot of potential, but really annoying." That line, reported by AI Leap (TI-Hüpe) project lead Ivo Visak in survey results published this month by ERR, is the most encouraging thing one could say about a government-built education tool. Estonia set out to make an AI that refuses to do students' homework for them. By that measure, the annoyance is proof of success.
The model is supposed to push back
ITI is a ChatGPT-based application built specifically for Estonian schools that, unlike consumer chatbots, declines to hand over finished answers. It asks questions back, prompts students to plan their own approach, and pushes them toward their own conclusions — a Socratic design Euronews describes as deliberately "technorealistic." The numbers Visak shared are modest but real: roughly 12,000 student accounts, with about 4,000 unique weekly active users. He sorts users into three groups — those deliberately avoiding AI, those who just want easier schoolwork, and a deliberate cohort using the tool to grapple with hard material. It is the second group that finds ITI most irritating, because the tool was engineered precisely to frustrate the shortcut.
That distinction matters. The animating fear behind AI Leap is not that students will use AI; surveys cited by Euronews suggest between 64 and 90 percent of Estonian students already were before the program launched. The fear is intellectual atrophy. As Visak framed it, "the question now is what we can do to ensure that people do not fall behind intellectually." A tool that simply completes assignments accelerates exactly the dependency educators worry about. One that withholds answers — at the cost of being annoying — is at least aimed at the right target.
A new chapter in a 30-year digital-state strategy
AI Leap is not a standalone gadget. It is the latest expression of the same institutional logic that produced Estonia's renowned digital state. The program is an explicit descendant of Tiigrihüpe ("Tiger Leap"), the late-1990s initiative that wired Estonian schools for the internet and seeded a generation of engineers. The same country that built X-Road — the data-exchange backbone, operational since 2001, that now carries roughly 2.2 billion transactions a year across more than 3,000 e-services — is now treating classroom AI as public digital infrastructure rather than a procurement afterthought.
The governance structure reflects that seriousness. According to the official TI-Hüpe program and the EU's Eurydice network, AI Leap launched on 1 September 2025 with around 20,000 students in grades 10–11 and roughly 4,700 teachers, under a 50-50 public-private funding model led by the Ministry of Education and Research with backing from the President's Digital Advisory Board and entrepreneurs behind Skype and Wise. The tool is built with Estonian education researchers and the Institute of the Estonian Language, and answers in Estonian — a deliberate act of linguistic and digital sovereignty for a small language community.
Steelmanning the skeptics
There is a serious case against any of this. A state-built AI tutor raises legitimate questions: Should a government be in the business of shaping how children think through a model it co-designs? What happens to pluralism when one publicly endorsed tool, tuned to "Estonian communication norms," mediates millions of learning interactions? And practically, an earlier ERR report noted real risk that students simply ignore a state-flavored AI in favor of the frictionless commercial alternatives a click away. A tool only 4,000 students touch weekly is not yet a movement.
These concerns deserve weight. But Estonia's approach is, on balance, the proportionate one — and a useful counter-model to the regulatory instinct elsewhere to ban or wall off AI in schools. Several jurisdictions have reached for prohibitions; Estonia instead assumed adoption was inevitable (the 64–90 percent figure suggests it was right) and invested in shaping the tool rather than outlawing it. Crucially, ITI is voluntary, transparent in its pedagogy, and sits alongside, not in place of, teachers — the program funnels resources into 4,000-plus teachers' premium access and professional learning communities. That is regulation-by-building, not regulation-by-restriction.
The expansion test
The model now faces its real stress test. Visak says the program is seeking an additional €1 million from the ministry and has reached preliminary agreements to extend into vocational education, with basic schools and a teacher base potentially exceeding 60,000 next in line — many of whom, ERR notes, have criticized the initial focus on upper-secondary students. Scaling a deliberately demanding tool from self-selecting high-schoolers to the entire system is far harder than launching it. A tool that 12,000 motivated teenagers find "annoying but worth it" may simply be abandoned by younger or more reluctant cohorts.
That is the wager worth watching. Estonia is betting that public digital infrastructure — the same philosophy that made a country of 1.3 million the world's most government-digitized society — can be extended to the hardest problem AI poses to education: keeping humans doing the thinking. The early data is thin and the verdict is grudging. But "annoying but full of potential" is precisely the review a well-designed learning tool should get. The alternative — an AI students love because it does the work for them — is the outcome every serious educator is trying to avoid.