Type in a topic — "cellular respiration," "intro to Python," "the French Revolution" — and Flow returns a structured course: modules, lessons, flashcards, and a quiz, ready to study from. Here's roughly how that works, and where it falls short.
Structure before content
The first pass doesn't generate lesson text — it generates an outline. Breaking a topic into modules and ordering them sensibly is a different problem than writing good explanations, and getting the structure wrong is much more disruptive to a learner than an individual lesson being slightly off. We generate and review the outline first, then fill it in.
Flashcards are derived, not separate
Flashcards aren't generated independently from the course — they're extracted from the same lesson content, which keeps terminology and emphasis consistent. If a lesson defines a term one way, the flashcard for that term matches it.
Quizzes target the outline, not just the text
Quiz questions are written against the module's learning objectives rather than purely against the generated text. This matters because it's easy for a model to write a quiz question that's technically answerable from the lesson but doesn't actually test understanding of the concept the module was meant to teach.
Where a human still matters
Generated courses are a strong starting point, not a finished product. We're explicit about this in the product: every generated course is editable, and we'd rather a learner adjust a lesson than trust something that reads confidently but is subtly wrong. For niche or fast-moving topics especially, review before you rely on it.
What's next
We're working on letting learners point Flow at their own material — lecture notes, a textbook chapter, a set of slides — and generate a course grounded in that source rather than general knowledge. That's the next big shift: from "teach me about X" to "teach me this specific material."
