Can an AI learning coach help children become more independent learners without taking over the thinking and struggling that learning requires? This question has shaped the development of enduri, which has now launched after three years of work.
I have spent much of my professional life trying to understand how children learn. As a teacher, education journalist and mother of four, one conviction has only grown stronger: Learning is deeply human. It involves curiosity, frustration, emotions, relationships, mistakes, motivation and the satisfaction of finally understanding something.
So when I started building enduri, I faced a contradiction. I wanted to use artificial intelligence to help children become better learners. But how could I do that without allowing AI to take over the very process I wanted to strengthen? And more fundamentally: Should I?
Can AI support learning without removing the struggle?
AI is exceptionally good at providing answers. But providing an answer is not the same as helping someone learn. Sometimes learning means struggling with a question. Trying something that doesn’t work. Going back. Asking another question. Discovering that what seemed impossible five minutes ago is suddenly within reach.
What happens when AI removes that struggle? That question has shaped many of our decisions at enduri. We had to rethink what an AI learning coach should actually do. Should it explain? When should it ask instead? How much help is too much? When should it step aside?These aren’t just technical questions. They are educational ones.
Building an AI learning coach that doesn’t give answers
With enduri, we decided to put the learner, not the technology, at the centre. Children aged 9–12 begin by discovering more about themselves as learners: their strengths, motivations, interests and needs. They explore practical learning strategies, set goals and reflect on their progress. And they meet Eni, an AI learning coach designed not to give them the answers, but to help them find their own way forward.
That sounds simple. In practice, it has been anything but.
Because building AI that gives answers is one thing. Building AI that knows when not to give an answer is another. And there is another challenge: Learning is individual. A question that helps one child move forward may leave another completely stuck. Supporting a learner means recognising that difference without making the decisions for them.
Three years later: enduri has launched
After three years of development, testing, discussions and difficult decisions, enduri is live. I am excited. But I’m also aware that launching a platform doesn’t prove that it works. The real questions begin now. Will children become more confident in tackling difficult tasks? Will they discover learning strategies they can use without enduri? Will they become more independent, rather than more dependent on technology?
That’s what we need to find out. I haven’t changed my mind about learning being human. If anything, building enduri has made that conviction stronger. The challenge was never to make AI do more of the learning. It was to find out whether AI could help children do more of it themselves.