Without friction, is there still learning?
This week, three applications tested approach learning in their own way. ChickyTutor makes us speak a language, then its avatar points out our mistakes and invites us to reformulate. In guided mode, Teachoo doesn’t give the answer: it asks questions and distributes clues when we get stuck. With Tomo, the course is built over the course of our answers to the quizzes and their correction, with immediate feedback. Three methods, the same bias: the refusal of the ready-made answer.
But not all friction is equal. The one from Teachoo and ChickyTutor is about reflection: you have to search, reformulate. Tomo adds another, on consumption: in the free version, “sparks” to spend and five lives that are renewed at a rate of one every two hours. These lives limit the time spent on the app (or bring frustration to get us to the paid plan…), not what we let it do.
What ultimately counts is not the amount of effort, but the moment in which it is placed. Philippa Hardman, whose synthesis relayed by Mickaël Bertrand feeds part of this focus, reports on an experiment conducted with 1,176 students (Ates, 2026). Those whose AI criticized each draft without giving them time to judge it themselves improved their texts the fastest. A few weeks later, without the AI, they obtained the worst results, while two minutes of self-assessment before the return of the AI preserved what they had learned.
However, the friction must not be circumvented with a click, which is what our brain, whether “efficient” or “lazy”, would do at the first opportunity. Teachoo also offers an option where the generated text gives the complete solution without asking the slightest question, as a majority of AI assistants do. However, among the students observed by Cui and his colleagues, a third of the uses switch to replacing reflection under the pressure of a grade or an imposed deadline, compared to 5.3% when they explore out of interest. And the phenomenon is not only academic: in front of lawyers and engineers, Marie Lacroix reminded us that everyone feels more effective with AI, without collective performance following.
It remains to be seen who resolves the friction. At school, it is often the teacher. In a second-grade class in Australia, a teacher taught a four-step method: querying the AI, checking on a search engine, asking a question again, rephrasing in his own words. In this class, researchers noted that the disposition to doubt was one of the conditions for successful questioning (Tang et al., 2026, study relayed by Sydologie in the readings). However, among adults, tutoring is often no longer present. Master’s and doctoral students at the University of Hong Kong, who are heavy users of AI, have set themselves a boundary: to be supported in their thinking rather than to outsource it (Dai and Chan, 2026, also cited by Philippa Hardman). But this position of collaborator remains rare: in the Cui corpus, it appears in only 7.5% of conversations.
When the tutoring is no longer there, no one solves the friction for us. A tool can offer it, the engineering of a training course or an employer may require it, but most often it is only us, and a response generated is just a click away. Without friction, however, we do not learn, we produce or have a response produced. Learning is going through steps, trials, mistakes that we correct, with a difficulty present enough to make us think but not to the point of discouraging when we are already struggling.
The question remains: do we know whether AI has helped us learn or only get an answer?