All three parts
Three parts, in order
Knowing AI, then Working It Out, then Building with AI. A comes first because a child who does not yet know the machine can be confidently wrong reads every answer in Part B as if an adult had said it.
Teacher-vetted project learning
AI literacy, critical thinking, life skills, and a digital portfolio
— with an interactive offline AI, for children 7 to 11.
All three parts
Knowing AI, then Working It Out, then Building with AI. A comes first because a child who does not yet know the machine can be confidently wrong reads every answer in Part B as if an adult had said it.
Part A · Knowing AI
What AI is, where it learned all this, where it goes wrong, and when not to use it at all. It is the part that makes the other two safe to do.
Part B · Working It Out
One idea, one right answer, five minutes. Any lesson that hands over a rule also hands over one case where that rule breaks.
Part C · Building with AI
Steps, something that physically exists at the end, and a real person who has to use it. This is the format that leaves the screen.
All three parts
Your child writes down a choice and a reason before the page will show what happened next. It is the only thing this site ever locks.
Part C · Building with AI
Named work with the failures left in. It stays with your child's account, not with the season they made it in.
All three parts
Part A reads from seven, Parts B and C from eight. A younger child starts with the part that never rotates, and grows into the other two.
Part B · Working It Out
Anyone can get an answer for free now. What is scarce is a child who can judge one — and half of this part happens with nothing switched on at all.
All three parts
Paper, a ruler, a kitchen scale, a timer. Projects use what is already in the house, and the materials are listed on the front of the task.
Part C · Building with AI
A problem with no steps. Your child decides what to make and what counts as done, and gets pulled into measuring, asking and working out things nobody set for them.
Part A · Knowing AI
Part A never rotates. There is no such thing as this season's Part A — whichever way you buy, the whole of it is on the shelf.
Part B · Working It Out
Ask the AI for a number you can measure, then go and measure it. Every one of these has a branch for the times the machine turns out to be right.
All three parts
A season is ninety days, four a year. Parts B and C get a new batch of topics with each one; what changes is the topics, never the shape.
Part C · Building with AI
Confidence comes from finishing something real, not from being told you did well. Your child picks a real problem and gets further with it than they could alone.
Part B · Working It Out
A choice and a reason, one sitting, nothing to make. Some of them have no right answer at all; some have a catch buried in them.
All three parts
From you, not only from your child. The parent script — what to say, what to stay out of, when to let it fail — is part of the product, not an extra.
Part C · Building with AI
In the long projects the failure is written in. Step two is the step that breaks, so that step three can be: now you know why.
All three parts
Inside a part there is no order at all. No streak to keep and nothing to fall behind on — start anywhere, stop anytime, pick it up a fortnight later.
Part B · Working It Out
The mirror of catching it out: here is a finished thing — how was it made? Then take one apart, or make one, and find out.
All three parts
It moves outward as the work gets realer. The course decides a Session is finished, your child decides a Mission is, and whoever has to use the Project decides that one.
Part A
Always included · Ages 7 to 11
AI literacy, and the safe way for children to use AI. This part comes first because a child who does not know it can be wrong hears every answer as a grown-up talking. They can practise on our offline AI, where what they say or type never leaves your child's own device.
Where it learned all this, the three rules that go on your wall, and the first question your child asks it for real.
No transformers, no diagrams. It learned from an enormous amount of what people have already written, which is why it sounds so certain — and why it has no way of knowing when it is wrong. This is also where a fact stops meaning “true” and starts meaning “something you can go and check”.
Never give it your name, your address or your school. What it says is not what a grown-up says. If something feels off, go and find a real person. Your child writes the three rules out in their own words, and everyone in the house signs the bottom.
Before the answer appears, your child writes down what they think it will say. The page will not move on until they have. Sometimes it turns out to be right — that branch is written in too, because “the AI is always wrong” is just as useless a habit as believing it.
Ask it for a number. Then go and measure that number yourself, with something from the kitchen drawer.
Not “is this a good idea” — a number. How much weight, how many centimetres, how long. The AI here is not an assistant and not a tutor. It is a source that hands you a figure, and the figure is the whole point.
A ruler, a kitchen scale, a timer, a stack of books. Your child does the measuring — not a search, not a second opinion from another model. This is the one section where an adult needs to be in the room the whole time.
Every catch in this course has a branch where the machine turns out to be correct. What we are marking is whether your child went and measured, not whether they caught it out.
The parent chapter. What a record is for, what happens when you press publish, and how to photograph work without photographing a child.
Credentials are turning into signals and work is turning into evidence. We are not going to tell you this gets anyone into a school — that is not ours to promise. We will tell you what a record with the failures left in is actually worth.
Local, sent to one person, public — and what each one really means. Then the practical half: photograph the finished thing, the hands doing the work, the object in use. Not the child.
Step by step, with screenshots. Nothing syncs on its own and there is no “back up everything” box — each piece goes in because you put it there.
Optional, and one piece at a time. What we check before it goes up, how long that takes, how to take it down again — and the honest part: taking it down stops new views, it does not undo the ones already had.
Part B
Rotates every season · Ages 8 to 11
Anyone can get an answer from AI for free now. What is scarce is a child who can judge one for themselves — and that is not a job a machine takes over but a skill that can be trained. Half of this part happens with nothing switched on at all.
A fact is not a thing that is true. It is a thing you can go and find out.
Three words children use as if they were the same word. The line this lesson installs is not true-versus-false — it is checkable-versus-not, and it is the line every catch later in the course depends on.
All three sound equally confident. Your child picks one and writes why, and the page will not show what happens next until they have.
A claim with no number in it cannot be checked by anyone. The work is rewriting it until it has one, then finding out where that number would come from.
Ask the AI how much a folded sheet will hold. Then stack books on it until you find out.
Before measuring anything, put a number on your own confidence. Children who never do this cannot tell the difference between being surprised and being wrong.
A choice and a reason, then the answer. Both folds are plausible. The reason is what we are after.
Paper and a stack of books, nothing bought. Ask for a figure, then load the thing until it gives way. There is a branch here for the times the figure was close.
One example is enough to break a rule. Finding that example is the whole skill.
Every lesson in this course that hands over a rule also asks the child to go and find one case where the rule breaks. This is the lesson that teaches them what they are looking for.
Most children say no. A balloon and a scale say otherwise, and the gap between those two is the lesson.
Designing the check is the work here, not performing it. If your child cannot describe how they would measure something, they cannot tell whether a number about it is nonsense.
Someone wrote that sentence, and they wrote it for a reason. What was the reason?
Not suspicion of everything. A single habit: before deciding whether something is true, notice who benefits if you believe it.
Deliberately from a different subject than the lesson. Children who only ever see motive in adverts learn to look for it in adverts and nowhere else.
No lie anywhere in it. This is the harder case, and the one children meet daily — a set of true sentences arranged to leave a false impression.
And the lesson underneath it: some questions have no right answer, so you cannot catch an AI out on them.
Taste is not a fact and not a guess. Knowing which questions cannot be settled by measuring is exactly as important as knowing how to measure, and this is the only lesson in the course that teaches it directly.
From living things rather than design, and deliberately so: the answer arrives when someone else has already done the waiting.
The ranking is not the answer. The stated criterion is. Two children can rank them in opposite orders and both be finished correctly.
A lot of people saying a thing, and somebody having measured it, are two different things.
The habit is one question: has anyone actually gone and checked, or does everyone just know? Children can tell those apart once they have been shown the difference once.
One says everybody knows it. One says somebody measured it. Your child picks, writes why, and then finds out which of the two could be settled today.
The quickest way to school, the coldest room, the longest queue. Pick one thing the household simply knows, and go and find out.
Three children of different heights want to watch the game. Is one box each fair?
Everyone gets one box. The tallest can already see and the shortest still cannot. Your child works out what went wrong before anyone uses the word fair.
The same amount each, or by who needs it. Neither is always right, and the reason your child gives is the whole exercise.
It has records and no idea who is in the room. No answer key, and the script says so in the lines the grown-up reads out.
Write down what a friend does. Then sort the list into what it can do and what it cannot.
Not a lesson about machines. A list of what friends do, sorted into two piles, and the sorting is something a child can check for themselves.
A choice and a reason. Most children pick the one who agrees, then hear themselves say why.
Ends on a question, not an answer: if one day it really could feel something, would you change your mind? Your child's answer goes home with them.
Twenty endings, written by a machine. Pick three, make one better ending out of them. Who wrote that?
Selecting is not copying. This lesson makes the difference visible by having your child do both and compare what was left in their hands each time.
No right order. The criterion your child names is the answer, and two children can disagree and both be finished.
The making is real and so is the question after it: what has to be true about a piece of work for it to be yours?
Somebody chose where to stand, what to point at, and what to leave out of the picture.
Not about fakes. Every photograph has a person behind it who chose a place to stand, and that choice is something a child can find and name.
Same object, two pictures, one choice and one reason. Then a look at where the camera was each time.
Nothing bought and nothing faked. Your child changes where they stand and what is in the frame, then says exactly what they changed.
Part C
Rotates every season · Ages 8 to 11
Confidence comes from finishing something real, not from being told you did well. Your child picks real world problems, and uses AI skills learnt from the course to get further with it than they could alone. Then they are real projects in their digital portfolio for them to be proud of.
Two steps, paper only, and a person at the end who has to actually put something on it.
Your child sets the target before building anything, and writes it down. Everything after this is measured against their number, not ours.
Paper and whatever is already in the house. If a project needs a shopping trip, nobody does it — so no project here needs one.
Not show it to. Use it. The person it goes to is the one who decides whether it was finished.
Three steps, and the second one is written to collapse — so that the third can be “now you know why”.
Again the child sets the bar, and again in writing. Using as little material as possible is the constraint that makes this hard.
It will break when it is full. That is not bad luck — it is written into the step. A project the course has designed cannot go wrong on its own, so failure has to be put there on purpose.
Rebuild it, having felt the thing give way. This is the only place in the whole course where consequence is guaranteed rather than hoped for.
This season's main mission. No steps at all — your child decides what to make and what counts as it having worked.
It cannot cost anything. The person has to actually use it — not something you decided they needed. And then your child adds a third constraint of their own.
There are no steps and we never say whether it worked. Copying the question back is not an answer; defining “finished” is the entire exercise.
Not us, not a badge, not a completion bar. The person it was for. Every household answers this mission differently, which is why it is the one the season's wall is built around.
One corner of your house that you see every day and quietly dislike. Change it.
Before any idea about what to do with it. Children who pick the idea first end up looking for a corner that fits the idea.
And why that. This is the mission where the reason matters more than the result, because there is no measuring your way out of it.
Someone in the house will prefer it the way it was. That is not a failure condition — it is the point, and it is the only mission this season where it can happen.
One price, both subscription plans, all three parts
$39/ monthcharged monthly, cancel any time
$119/ six monthsthat is just under $20 a month, cancel any time
$39one month
Gift a month$119six months
Gift six monthsFree refund within 30 days — no reason needed, one button, gifts included. You can read the first chapter before you buy anything. Prices are in US dollars, and checkout is not open yet — these are the plans and prices it will have.
An eight-year-old does not run a project alone. The parent script — what to say, what to stay out of, when to let it fail — is part of the product, not an extra.
Part A asks for more. In the chapter where your child measures things, you are in the room the whole time. If nobody in the house has that hour and a half, this is not for you.
It makes mistakes, invents things, and sometimes cannot answer at all. Catching it is the exercise: ask it for a number you can measure, then go and measure it.
Every task also branches for the times it turns out to be right. “The AI is always wrong” is as useless a lesson as believing it.
Paper, books, a ruler, a kitchen scale, a timer. Projects use what is already in the house, and the materials are listed on the front of the task.
Anything with a browser. The AI downloads once and then runs on the device itself — after that it works with the internet off, and nothing your child types is ever sent anywhere. Reading pages aloud also needs a GPU the browser can reach — most laptops, desktops and recent iPads. Without it nothing else changes: every item can be finished with it switched off.
Ninety days, four a year. Parts B and C change with each one; Part A never changes. Six months covers two seasons, monthly covers whichever one is open.
We strongly suggest it. A child who does not yet know the machine can be confidently wrong will read every AI answer in Part B as if an adult had said it.
Inside a part there is no order at all. The only thing the site ever locks: your child cannot see what happened next until they have written down what they think and why.
Then have the page read itself. It runs inside the browser, so nothing your child is reading gets sent anywhere.
Reading aloud · on the device
The voice downloads once, then runs on your own machine. Nothing your child is reading is sent anywhere to be spoken. One small record does travel, so that the portfolio survives a new laptop: which item was finished, and when. No writing, no photos, no conversation.
The part of the course that answers a child’s questions works the same way — a bigger download, once, and then it runs on the machine too. The privacy page says what does leave: the work you choose to save.
A few lines from a lesson about how much weight a folded sheet of paper will hold.
The first press downloads about 311 MB of voice model, once, from huggingface.co — the one connection this layer makes. After that it works with the network switched off.
It runs on laptops, desktops, iPads and phones — anywhere the browser can reach a GPU.
None of it is required. Every chapter is written to be finished with it switched off.
Yes, either length. A gift is charged once — $39 for one month, $119 for six — and it never renews. You pay, we send a code, and the parent receiving it opens the account.
Paying is not the same as the account being live. The child is under 13, so consent comes from the parent who will actually be there. You will not see anything about the child afterwards.
A certificate says somebody assessed your child once. A record shows what they actually did — including the parts that did not work, which is the half nobody keeps and the half that is hardest to fake.
We are not going to tell you this gets anyone into a school; that is not ours to promise. What we will say is that every piece in the portfolio carries a line your child wrote about why it mattered to them, and that line is usually worth more than the thing itself.
What your child types, writes and photographs stays on their device. The pieces you deliberately save go into your account, and one small record travels on its own so the portfolio survives a new laptop: which item was finished, and when. No writing, no photos.
We keep that portfolio for as long as you have an account — it goes when you delete the account, and you can export or delete any of it yourself at any time.
Why?
Answers, code and analysis at close to expert level now cost almost nothing. Being willing to put your name on one did not get cheaper.
The first three teach what to know. We teach what to do when the machine already knows it.
Whether the answer is any good — and how to build something real out of it.
We are not teaching children to use AI — we are using AI as sparring practice, until they no longer have to take its word for anything.
Where to start
Everything the course is, what it costs, and what it asks of an adult is on this page. Checkout is not open yet.