The learning loop
Serious learning products create retrieval, feedback, spacing, interleaving, motivation, and performance practice. The learner has to generate an answer, phrase, proof step, classification, prediction, or decision from memory.
Brilliant: productive struggle with guardrails
Brilliant's product shape is useful because the learner interacts with the concept early. The lesson asks the user to predict, manipulate, infer, and correct, with enough scaffolding to keep the struggle productive.
Duolingo: retrieval plus habit
Duolingo turns frequent retrieval into a daily habit. Its product insight is not only spaced review or quick feedback. It is making review feel like progress rather than punishment.
Math Academy: a knowledge model
Math Academy makes the machinery of personalization explicit: a domain model, a student model, prerequisite structure, review timing, and task selection. That is the difference between an adaptive learning system and a smart playlist.
Speak: conversation as performance practice
Speak points at a key principle for performance skills: simulate the real act early. Speaking requires generation under pressure, not only recognition on a screen.
What builders should steal
Make the learner act. Treat feedback as a core interface. Build a memory system. Make review feel like progress. Interleave enough to force choice. Constrain AI tutors with pedagogy, learner state, safety boundaries, and evaluation.
Source notes
This piece was checked against official/public materials from Brilliant, Duolingo, Math Academy, Speak, and cited learning-science papers on retrieval, distributed practice, active learning, and deliberate practice.