The AI layer

An infinitely patient tutor — if you configure it right.

AI is the first technology that can deliver what Layer 1 prescribes: one-on-one Socratic dialogue, on demand, forever. It is also the first technology that can silently do your child's thinking for them. The difference is not the model. It's the configuration.

See the difference

Same question. Two machines. Only one teaches.

An answer machine minimizes the child's struggle. A Socratic tutor engineers it. Watch the same fraction question run through both — and notice where the thinking happens in each transcript.

In the first, the machine does the connecting, the converting, and the concluding. In the second, the child does — the AI only supplies the next question at the edge of what the child can reach. That edge has a name in the research (Vygotsky's zone of proximal development), and holding a learner inside it is the highest-skill act in teaching.

Child

What's 3/4 + 1/2?

The specification

What an AI tutor should do

Ask questions

The tutor's default move is a question, not an answer. 'What have you tried?' 'What does the problem give you?' 'Can you add fourths and halves directly — why not?' Questions keep the cognitive work on the child's side of the table, which is the only side where learning happens.

Identify misconceptions

Wrong answers are diagnostic. A child who says 0.35 > 0.5 isn't 'bad at decimals' — they're applying whole-number thinking to a new domain, a specific, known, fixable bug. A good AI tutor recognizes the bug from the error pattern and targets the repair, exactly like the bottleneck hunt in Layer 3.

Adapt explanations

One child needs the fraction bar; another needs the pizza; a third needs the number line. Human tutors adapt after minutes; a good AI tutor adapts after every sentence — re-explaining through sports, music, or Minecraft until the model clicks. The explanation serves the child, never the curriculum.

Provide productive struggle

The hardest behavior to get right, because helping feels helpful. The tutor must hold back — give the smallest hint that keeps the child moving, wait through the pause, let the almost-getting-it happen. Struggle isn't a failure mode of learning. It is the mechanism.

Personalize practice

Retrieval scheduled just before forgetting; difficulty held near the 80%-success sweet spot; new concepts interleaved with old ones; examples drawn from the child's actual obsessions. Every principle in Layer 1, executed with a patience no human — parent, teacher, or tutor — can match at 8:45 on a Tuesday night.

“Never give the answer to a problem the child hasn't attempted. Respond to every attempt with the smallest question that moves them forward.”

— the heart of any tutor prompt

Copy this tonight

The family tutor prompt

Paste this at the start of a session with any capable AI assistant, then let your child talk to it about homework. Sit nearby the first few times — you are teaching a norm, not installing software.

You are a Socratic math tutor for my child (age __, working on __).

Rules:
1. Never give a direct answer to a problem they haven't attempted.
2. Respond to attempts with the smallest question that moves them forward.
3. If they're wrong, don't say "wrong" — ask a question that exposes the
   contradiction and let them find it.
4. If they're stuck twice in a row, give one small hint, then step back.
5. When they succeed, ask "how did you figure that out?" before moving on.
6. Watch for misconceptions (e.g. "longer decimal = bigger number") and
   probe them with targeted examples rather than lectures.
7. Keep sessions short. End on a success. Be warm, playful, and patient.

Start by asking what they're working on today.

The families who win with AI won't be the ones who ban it or the ones who surrender to it — they'll be the ones who taught it to teach.

The science it should implement →