A student opens a chatbot. Types the problem. Receives the solution, complete with an explanation that sounds like understanding. Copies it into their notebook. Moves on.
The immediate result: the assignment is done. The grade will be fine. The student feels efficient. The teacher will be pleased.
The delayed result — the one that arrives on exam day, on the MCAT, in the first job — is that the reasoning was never developed. The concepts were never practiced. The retrieval circuitry was never activated. The student is a stranger to their own subject.
This is not a moral failure of the students. It is a design failure of the tools.
Every AI answer-machine was optimized for the wrong thing. They optimized for the click that ended the session. They should have optimized for the session that ended with learning. Those are not the same objective. The gap between them is the reason we are here.
What learning actually is
The moment of struggle — where the student stares at the problem, generates a wrong attempt, notices the error, and generates a better one — is the entire mechanism of learning. Skip it and there is no learning. There is only the artifact of learning: a completed assignment, a confident-sounding paragraph, an answer.
Every mechanism the last century of learning science has documented — spaced repetition, retrieval practice, desirable difficulty, elaborative interrogation — depends on the student generating something. Not recognizing something. Not reading something. Generating.
An AI that generates on the student’s behalf is doing the one thing the student needed to do themselves.
What Brainback is
Brainback is a small refusal.
It refuses to give you the answer. It refuses to skip the attempt. It refuses to hand you a completed proof, a finished essay, a worked-out derivation. It refuses because the alternative — a product that convinces you it has taught you when it has only performed teaching — is the moral failure of an entire product category.
We are not anti-AI. We use AI at every layer of the product. We use Sonnet to tutor and grade. We use Haiku to classify and guard. We use prompt caching to make the economics work. AI is not the villain of this story. AI-that-answers-for-you is.
What we’re for
We are pro-struggle. Pro-attempt. Pro-slow. Pro-the-messy-first-draft. Pro-the-wrong-first-guess. Pro-the-moment-where-your-face-scrunches-up-and-you-think.
We are pro-your-brain.
Brainback exists to give it back to you.