Why we are building Thincker
AI can think for your students.Ours won't.
A tool that hands a student the answer has taught them one thing: that the answer was always available without them. Thincker exists to make the student do the difficult part — and to give the people who design their courses the means to be sure the difficult part is actually there.
Cognitive surrender is real, and it is quiet.
This generation of students has grown up with AI ready to think for them, and it shows: professors and researchers alike are watching critical-thinking and problem-solving skills erode as the reliance deepens. That’s the reason I started Thincker — not to keep students away from AI, but to make sure it strengthens their thinking muscle instead of letting it atrophy, and takes their thinking further than they’d get on their own.
That is not a failure of this generation of students. It is what any of us would do when something else will reliably think for us. Researchers have a name for it: cognitive offloading — outsourcing a mental task instead of doing it yourself. Gerlich’s 2025 study of 666 people linked frequent AI use to weaker critical-thinking scores, with offloading as the likely mechanism. Correlation, not proof — he says so, and so do we. Read the study.
The damage doesn’t announce itself, either. Students using an AI tutor often do better while they have it — that is exactly what makes it hard to catch. In one high-school maths trial, an unrestricted GPT-4 tutor lifted practice scores, then the same students scored worse on the exam once it was taken away. The assistance had replaced the practice, not supported it.
An answer is not an education. A tool that ends the thinking is not a teacher, however fluent it sounds.
What a great tutor does isn’t a mystery.
And it isn’t simply attention. A good tutor knows which idea you are missing, not which page you are on. They ask before they tell. They catch the confident wrong answer. They keep returning to an idea until it actually holds — not just for the next quiz, but for good. Each of those is a studied, documented mechanism — retrieval practice, spacing, misconception-first instruction, calibrated questioning.
2σ
Benjamin Bloom put a number on it: students taught one-to-one scored about two standard deviations above students in a conventional classroom — the average tutored kid outperformed most of the untutored ones. He called it the 2 Sigma Problem — because nobody could afford to give every child a tutor.
That was never a limit of teaching. It was a limit of economics. The classroom is what we could pay for, not what we knew worked best.
Those mechanisms are what we’re building. Not a chat box.
Three things we build differently.
Every decision in the product resolves to one of these.
01
The default
Answersarguments
An answer ends the thinking. A good question starts it. Once a concept has been introduced, Thincker tests whether the student actually understood it — not whether they can recite it back.
02
The unit
Pagesconcepts
A course is not a pile of documents. It is a structure of ideas that depend on each other, and knowing which idea a student is actually stuck on is the whole game.
03
The measure
Completionmastery
Finishing a module proves attendance. We care whether the understanding survives a week later, and whether the assessment tests the thinking it claims to.
Two people have to think harder.
For people who build courses
A thinking partner, not a text generator.
- Pressure-tests the reasoning behind a lesson, not just its prose
- Shows what each part of your course actually teaches — and what it only appears to
- Surfaces where your material and your assessments have drifted apart
For students
Something that refuses to do the work for you.
- Asks before it tells, so the first move is always the student’s
- Meets the student at the idea they are stuck on, not the page they are on
- Brings ideas back before they fade, rather than after the exam
The standard we hold ourselves to.
Generating plausible course material isn’t hard anymore. A tool that stops there isn’t worth building. The real question is whether the material teaches what it claims, whether it demands real thinking or just recall dressed up as thinking, and whether a student who finishes it actually understands anything.
We would rather build the thing that makes a student capable without us than the thing they cannot work without.
Sources
- Bloom, B. S. (1984). The 2 Sigma Problem — the one-to-one tutoring result, and why it stayed out of reach
- Bastani et al. — Generative AI Without Guardrails Can Harm Learning — randomised high-school maths trial: better with the tutor, worse once it was removed
- Gerlich, M. (2025). AI Tools in Society — Societies 15(1), 6 — AI use, cognitive offloading and critical thinking across 666 participants; correlational, and a correction was later published
- Fan et al. — Beware of Metacognitive Laziness — ChatGPT raised essay scores without improving knowledge gain or transfer
Thincker is invite-only while we build.
If you design courses and this is the argument you have been making to yourself, we would like to talk to you.