The workflow is the whole product. AI answer-machines optimize for the click that ends the session; Learn mode optimizes for the session that ends with learning. Those are not the same objective, and the difference shows up in the design.
The attempt-gate is the most user-facing rule. It’s also the one that gets the loudest complaints for the first three days and the loudest praise for the next three weeks. A cognitive load that felt like friction becomes a scaffold you count on. The moment the friction pays back is the moment the product starts working.
What Learn mode is not
Learn mode is not a chat window with a “don’t give me the answer” system prompt. A system prompt is a stance; a stance drifts under pressure. Every rule in Learn mode is enforced by code that runs regardless of what the model was told.
It’s also not a homework helper in the copy-paste sense. It’s the tool you use when you want to actually understand what you’re doing. If you want a paragraph drafted for a topic you aren’t being graded on, Efficiency mode is the door to your right.
What Learn mode replaces
The last two decades of undergraduate education have run on office hours + TA sections + the occasional heroic study group. Those things are great and Learn mode does not aim to replace them. What it replaces is the middle-of-the-night alone-with-a-problem-set failure state, which used to end with a copy-paste from ChatGPT and now ends with a Socratic exchange that either produces the answer in your head or produces a clear question you can walk into office hours with.
The night-before-the-exam “I don’t know how to think about this” is where the compounding shows up. If you’ve used Learn mode for a semester, the tools are yours; if you’ve used an answer-machine, they aren’t.