In the Czech Republic, an NGO bets on AI to support student learning

The French government just announced a new reform: two hours of weekly AI literacy training starting in the second year of high school, beginning in the 2027 school year. Over in the Czech Republic, the NGO AI dětem is betting that AI can do something similar but for a different skill: helping students learn to question, reason, and think for themselves.

Today we're taking a closer look at Tiny, a suite of chatbots built to help a generation of students become more independent and confident learners in the classroom.

We spoke with Kateřina Çakın and Artem Redchych, CEO and CTO of Tiny, about the progress they've made and the obstacles they've run into; all while keeping in mind the questions that matter most to us: what does a genuinely useful application of AI look like for a nonprofit's mission? And where does it fall short?
‍

AI dětem, in brief. Founded in 2022, this Czech NGO helps schools prepare new generations of students for a rapidly changing world. Its work rests on three pillars: producing educational content, training teachers and students, and developing digital tools for schools.

Tiny, in brief. Tiny is AI dětem's technology solution. It brings together three applications (for administration, teachers, and students). On the student side, it relies on 11 educational chatbots, the TinyBots. Today, AI dětem works with more than 333 schools and has reached more than 12,000 students aged over 13 with Tiny. The ambition: to reach 10% of Czech students, or 350,000, by 2028.

‍

‍

The underlying problem: AI use that undermines learning
‍

In the Czech Republic, as everywhere else, most students openly admit to using AI every day. That alone is forcing schools to rethink how they teach and to start training students in using AI responsibly. Because using AI to get an answer is one thing. Using it to ask better questions, test an idea, or push back on your own reasoning is something else entirely.

A recent study (Bastani et al., CMU/Oxford/MIT/UCLA, 2025) found that students who used an AI tool to solve problems actually progressed less than students who worked without one. And when the tool was taken away mid-exercise, the AI-assisted group gave up faster. Strip out the effort, and AI hollows out both the learning and the willingness to keep trying.

In a classroom of 30 kids with wildly different needs, one teacher simply can't be everywhere at once. Left unchecked, AI can actually widen that gap: students who already know how to use it well pull further ahead, while those leaning on it as a shortcut stall out, often without even realizing it.
‍

Chatbots designed to encourage critical thinking
‍

AI dětem built Tiny, an educational app made up of 11 chatbots, the TinyBots, each designed for a specific classroom purpose: reviewing material, exploring a new topic, brainstorming, or role-playing a historical figure. Teachers pick the bot that fits their lesson, launch the session, and students work through it individually on their own screens.

The TinyBots don't hand out answers, they ask questions instead. A student who wants information has to dig for it, put it into words, make a choice. The whole system is built around Bloom's taxonomy, a well-known framework that breaks thinking into six levels: remembering, understanding, applying, analyzing, evaluating, and creating. Each TinyBot is calibrated to one of these levels.

Learning scientists have a name for this kind of friction: desirable difficulties (Robert and Elizabeth Bjork). The harder a student works to reason their way to an answer, the more likely they are to actually remember it.
‍

Keeping humans in the loop: the teacher remains in control
‍

While a class works through the TinyBots, the teacher watches it all unfold on a live dashboard: every student's conversation in real time, summaries of where they got stuck, and alerts whenever someone drifts off topic.

One rule was non-negotiable from day one: AI doesn't grade students. Its job is to surface what a student covered, where they struggled, and where they wandered off track. It's the teacher who interprets all of that and decides what happens next : a grade, a comment, extra help, whatever the moment calls for.

Picture a teacher prepping a lesson on World War II with a TinyBot. The bot flags that a third of the class only asked factual questions like dates or names, without ever digging into causes or consequences. Armed with that, the teacher can spend ten minutes addressing exactly that gap with the whole class.
‍

‍

‍

What's standing in the way?
‍

Technically, Tiny holds up well. It was stress-tested in real conditions during the AI Olympics, which drew 10,000 sign-ups in just a few days and let the team catch and fix bugs at scale. The architecture is already built to support a mobile version, and the model keeps getting sharper thanks to two full-time staff combing through the 45,000-plus conversations the platform has generated so far.

The bigger obstacle is infrastructure. Czech public schools often deal with unreliable internet, and an app that depends on a solid connection can become useless fast once it hits the real world.

Tiny's beta testers tend to be early adopters who are naturally excited about innovation but most teachers are stretched thin and skeptical that digital tools actually improve learning outcomes. Growing concern about teen screen time and mental health only adds to the wariness toward putting yet another screen in front of students. To keep growing, AI dětem will need hard data proving its approach actually works.

Do students learn better with Tiny? Do they build a healthier, more independent relationship with AI? To find out, AI dětem is launching a study with Charles University comparing outcomes between students who use Tiny and those taking the same classes without it. The results should offer real answers about Tiny's actual impact on learning.
‍

Three key takeaways if you want to get started
‍

AI, and technology in general, isn't automatically the right answer to your beneficiaries' problems. But if you do want to explore it, here are three lessons from AI dětem's experience:

1. Start with the problem, not the tool. AI dětem didn't set out to "add AI to lessons." They started by identifying a gap: in a class of 30, no student really gets an individual voice. Before you build anything, ask yourself: when, exactly, are my beneficiaries on their own, without support, at a moment that actually matters?

2. For every AI feature you're considering, get clear on what the human does in parallel and what only a human can do. AI can observe, summarize, flag. It can't take responsibility for what it concludes. If you can't clearly articulate what role the human still plays, you're probably replacing them, not supporting them.

3. Design for friction, not against it. If your AI tool makes things too smooth, too instant, check whether you've accidentally removed something valuable. Effort, hesitation, not knowing the answer yet is often exactly where learning happens. Ask yourself: am I cutting out friction that's just annoying, or friction that's actually doing something useful for the people I'm trying to help?

‍

Tiny is a project supported by the forwa program, Europe's first program dedicated to helping 14 organizations tackle inequalities in access to education and employment by putting AI to work for their mission. Through mentorship, training, and a peer community, we help these organizations scale their impact by building AI-powered digital tools.

‍

Our Sources

- Desirable difficulties : Robert A. Bjork & Elizabeth L. Bjork, "Desirable difficulties in theory and practice", Journal of Applied Research in Memory and Cognition (2020)

- AI and cognitive reliance : Bastani et al. (CMU, Oxford, MIT, UCLA), arxiv.org/abs/2604.04721 (2025)

- Bloom's Taxonomy : Benjamin Bloom et al., Taxonomy of Educational Objectives (1956); revision by Anderson & Krathwohl (2001)

- The Anxious Generation (2024), Jonathan Haidt

‍