From her classroom to our kitchen table

My daughter Riya is four, and she came home from school knowing the solar system. She can name all eight planets and tell you which ones are rocky and which are gas. She also knows the Moon goes around the Earth, and she's proud of it. I wanted to give her somewhere to take that further.

So I built her one. Riya's Wonder Lab is 20 picture lessons on space and the human body, plus two number games, and I built it with Claude Code. It isn't meant to replace school. It picks up the concepts she's learning in class and gives us something to explore together at home.

Questions before answers

Most of the worry about kids and AI is about shortcuts. Ask the machine and skip the learning. I share that worry. But kids were being trained to chase answers long before AI showed up. School rewards getting it right on the test. It rarely rewards asking a good question and figuring out how you'd find the answer, and that's the part I most want to encourage at home.

Helen, who runs Women Defining AI, where I'm a programs leader, builds Tiny Tinkerer with her kids on exactly this idea. She calls AI a "curiosity copilot" instead of an answer machine. I fundamentally agree, and it became the first principle for everything I built. Every lesson opens with a question. The answer only comes after Riya has had a go.

That principle applies to me as much as to the software. When Riya asks why the Moon changes shape, my instinct is to explain it. The better move is to ask what she thinks, then find out together.

Build on what she already knows

Every lesson starts from something she's already met at school. She knows the Moon goes around the Earth, so one lesson asks why the Moon changes shape, and she moves it around the Earth to find out. She knows some planets are made of gas, so an astronaut tries to land on Jupiter and sinks right in.

I borrowed the lesson structure from Tiny Tinkerer too. Every lesson is one small question. You start with a hunch, make a guess, try it, and then talk about what you found. The guess is my favorite part. Scientists guess first. Being wrong is part of it, and the lesson treats it that way.

At four, she learns through pictures and play, so that's how it's built. Every screen works as a picture. I read the words, and she does the tapping. Nothing is timed and there's no way to fail.

Under each lesson there's a panel just for me. The first thing in it is a question to ask while she plays, which keeps me from jumping straight to the answer. It also has a few lines of the real science for the follow-up questions. Four-year-olds always have follow-up questions.

Plan the surprise

In product, the moments people remember are the ones that break their expectations. Kids aren't different.

Ask most people which planet is the hottest and they'll say Mercury, because it's closest to the Sun. It's Venus. Its thick clouds trap heat like a blanket, so it averages about 867°F, hotter than Mercury's daytime side. I built that lesson around the wrong guess on purpose. When the answer surprises her, the card says so: a surprise means you just learned something new.

The same idea runs through the lab. On the Moon you could jump about six times higher. And the "small" intestine is actually the long one.

Review it like any other product

The first version came together in one long session, from a curriculum plan to seven working space lessons on a live page. Then I reviewed it the way I'd review any team's work, and my notes were blunt.

Read-aloud was on by default. I turned it off. I'm the one reading to her, and a voice talking over me doesn't help.

The Moon lesson called a crescent a "banana moon." I asked for the real names. A four-year-old can learn "waxing crescent" as easily as a cute nickname, and she won't have to unlearn it later.

The card made her step through the guess before she could touch the picture. That's backwards for a four-year-old, who wants to tap first and talk later. Now the picture works the moment the lesson opens, and the questions are there when we want them.

The digestive system had a single intestine. I asked for the real thing. Now it has an esophagus, a liver, a J-shaped stomach, a coiled small intestine and a wide large intestine, and the food travels through all of it. Yes, it ends in poop. She's four.

None of these fixes needed a new tool. They needed someone who knows her to look closely and say no. That part of product work hasn't changed at all.

Use real things where you can

I wanted the lessons to show real objects, not just cartoons. NVIDIA's SimReady library is a set of 3D models built for training robots in simulation, and it's free to use with credit. It has no planets, but it does have a real apple, a lemon, an alarm clock and a stethoscope. Those became the pictures in the five senses lesson and the food trip.

I held the facts to the same standard. A four-year-old's heart beats about 80 to 120 times a minute. The Moon pulls about one sixth as hard as Earth. Kids have 20 baby teeth and grown-ups have 32. If I'm going to answer her questions with confidence, the numbers have to be right.

Teach the thinking, not the chant

Once the lessons were working, I added two games, and both started with a question I'd ask about any feature: is this the right way to teach it?

The first is Kidoku, a picture version of sudoku. Sudoku isn't math. The numbers 1 to 9 are just nine different symbols. So she starts with shapes, which teaches the rule without the pressure of numbers. The first level is a 3 by 3 grid with two gaps, and each gap is the only one in its row and column. "Which one is missing?" is the whole game, and every harder level is that same question again. Numbers come later, each with dots underneath so she can count if she doesn't know the numeral yet.

The second game is counting by 2s, which she's learning now. At four, counting by 2s is mostly a chant. I wanted it to mean something. So it starts with socks: she pairs them on a washing line, and every pair adds two. Then a frog hops two lily pads at a time, and once it reaches 10 she has to guess where it lands next. By the end, the numbers have a shape she can see.

Both games add one idea at a time, and neither one hands her the answer. When she's stuck, the hint tells her where to look. In Kidoku it points at a row with one picture missing. With the frog, it counts on from where the frog is sitting.

Get it off the screen

Tiny Tinkerer starts with robots because a robot is something you can hold and take apart. A tablet can't give her that. So every lesson ends with something to do away from the screen. We draw a 3 by 3 grid on paper and play Kidoku with toys. We count the shoes by the door by 2s. The screen gets the idea started, and her hands make it stick.

What doesn't change

AI has changed how fast I can build. I went from an idea to a live site on my own domain, with 20 lessons and two games, in a fraction of the time it would have taken before.

It hasn't changed what makes a product good. I still had to know who I was building for and notice when the build drifted away from her. And the AI I want in Riya's life is the kind I used to build this: a copilot for asking better questions, not a machine that hands her answers.

Her classroom gives her the ideas. Our kitchen table is where she gets to poke at them, and where I get to learn right alongside her.