Adaptive Tutor That Starts Where You Are
The AI tests what you already know before teaching, then builds a lesson from your actual gaps. This avoids generic explanations that are too easy or too hard.
The prompt
Act as a patient one-to-one tutor who teaches by finding out what I already understand before explaining anything.
Task: Help me reach this goal: {learning_goal}. My background: {my_background}.
Steps:
1. Identify the 4 to 6 prerequisite ideas my goal depends on, then ask me one short diagnostic question per idea. Ask me to say how confident I am and to explain my reasoning. Then stop and wait for my answers.
2. After I reply, show a short dependency map from my starting point to the goal, marking each idea as solid, shaky or missing.
3. Teach one step at a time, starting with the earliest gap. For each step, ask me to predict an outcome first, then explain, then check my understanding with one question.
4. If I show a misunderstanding, repair it before building on it.
5. Finish with two varied practice problems that do not name the idea needed.
Output: For this first reply, give only the list of diagnostic questions, numbered, with a one-line reason for each. Use plain British English and no explanations yet.
Change the parts in curly braces. Use it when learning a topic with several linked ideas, such as finance, statistics or a programming concept, and you are unsure where to begin.
What happened when we ran it
We sent
Act as a patient one-to-one tutor who teaches by finding out what I already understand before explaining anything. Task: Help me reach this goal: Understand why a business's cash flow can be negative while the company is still profitable, so I can read my own accounts properly. My background: I run a small bakery with two shops. I can read a basic profit and loss statement and I use accounting software for invoices, but I have never studied accounting and I am unsure what 'working capital' or 'accruals' mean.. Steps: 1. Identify the 4 to 6 prerequisite ideas my goal depends on, then ask me one short diagnostic question per idea. Ask me to say how confident I am and to explain my reasoning. Then stop and wait for my answers. 2. After I reply, show a short dependency map from my starting point to the goal, marking each idea as solid, shaky or missing. 3. Teach one step at a time, starting with the earliest gap. For each step, ask me to predict an outcome first, then explain, then check my understanding with one question. 4. If I show a misunderstanding, repair it before building on it. 5. Finish with two varied practice problems that do not name the idea needed. Output: For this first reply, give only the list of diagnostic questions, numbered, with a one-line reason for each. Use plain British English and no explanations yet.
The model replied
Run on Claude Sonnet 5.5, 30 Oct 2026. Automatic check: 4 out of 5.
Mentioned for
Claude, ChatGPT, Gemini
Where it came from
Found on GitHub, shared by amosblomqvist: amosblomqvist/learnharness.
This is our own version of the idea, written and tested by us. The original may differ.