Each student spends ten weeks on one project of their own: a small game or tool that runs in a web browser. The AI writes the code. The student's job is to say exactly what they want, look carefully at what came back, and decide whether it's right. That skill will outlast any particular AI tool.
Building with AI is two things: a ten-week class program, and the Studio, the software students build in. They work best together, but the Studio isn't tied to the course. Use my curriculum, adapt it, or teach your own.
My ten-week program: each student plans, builds, tests and finishes one project of their own, and learns six habits for working with AI along the way.
A building tool made for classrooms. Students describe what they want and the AI builds it as a working web page. It works with any subject and any curriculum.
The Studio isn't built around one curriculum. Run my ten-week course as written, change it, or design your own: a science unit where students build simulations, an English unit where they build interactive stories, or an elective that's entirely yours. The plan interview, the warm-up, partner testing and the AI review are switched on per class from the teacher console, so you use the parts that fit how you teach.
Students type what they want; the AI builds a working page, game or tool they can use straight away.
Each build is saved. A student can look back at any version and bring it back with one press.
A short interview turns a student's idea into a written plan, and the AI reads it on every build.
An “I'm stuck” button and clarifying questions point students back to their own thinking, never to a finished answer.
The AI lists what it thinks is broken against the student's plan; students test each claim and decide.
Classmates try each other's projects, and finished work can be published for families once the teacher approves.
Every request in the class, as it happens, with each student's project one click away.
Open and close the Studio, set the lesson phase, and cover every screen with “Eyes up” when you need attention.
If a student writes something worrying, the teacher is alerted within seconds and the record is kept for the school.
Students sign in with a project code. No names, emails or school accounts are ever entered.
Student pages run on their own address and cannot reach the internet at all.
Runs in a current browser on Chromebooks, iPads and laptops, with touch or keyboard.
Take the course and the Studio together with me teaching, or license the Studio and run it your way.
See the ways to run itEvery session follows the same loop. The Studio doesn't chat, answer questions or write essays, doesn't remember a student from one class to the next, and only opens while the class is in session.
What does it do, who is it for, and how will you know it works? The answers become the project's written plan, and the AI reads that plan every time the student builds.
The student describes a change, the AI builds it, and every version is kept so nothing is lost.
The student tests what came back, fixes what doesn't match, then watches a classmate use it without being told how it works.
One project carries through all ten weeks. The first five weeks each add a habit, the next three put them to work on a bigger build, and the last two finish the project and show it. Every habit is about the student's judgment, not the AI's.
What generative AI is, and why it can sound certain and still be wrong. It only knows what you tell it; everything else, it guesses. Students write their three-question plan and try their first ideas.
Each student commits to one project. They learn to report a problem the AI can act on: "I did this, I expected this, instead this happened."
An AI reviews each project against the student's own plan and lists what it thinks is broken. The students' job is to check the AI: try each item themselves and decide whether it's really broken or the AI is wrong. When an AI says "done", that's a claim, not a fact.
Then you know which change caused what. Students fix their real problems one at a time and check again after each.
A classmate plays the project without being told how it works. The maker watches, notes what confused them, and fixes it.
Bigger features, using every habit so far, with a partner test each week. Students keep their plan up to date as the project grows.
Polish, then write how they made it: what they asked for, what the AI got wrong, and what they'd change next.
Families and classmates play the finished projects, and each student explains their own.
Each one was made the way a student makes one: answer the plan questions, build, then describe what's wrong and build again. Nobody edited the code by hand.
Built by the instructor from student-style requests, for this page. Student work is never shown here.
The single most important decision is what the Studio isn't: it is not a chatbot. There is no conversation to have, no character to talk to, and nothing to ask it except to build a page. Everything below sits behind that.
Building is possible only while the class is in session with the teacher present. Outside class, students can look at what they made and nothing else.
Students sign in with a project code, never a name or school account. The list matching codes to students stays with the teacher, on paper.
Every request is checked before anything is built, and every page is checked again before the student sees it. Anything uncertain is held back.
Pages students build run on their own address and cannot reach the internet at all: no outside images, scripts or trackers, and no way to send anything anywhere.
If a student types something that suggests they may need help, the Studio doesn't try to counsel them. It tells them their teacher has been told, and alerts the teacher within seconds.
A live console shows the teacher every request in their class as it happens. Records are deleted on a fixed schedule, except safeguarding records, which a school administrator releases.
The full privacy and data-retention policy: the Studio's privacy page.
After school, on your campus, for schools within about 15 minutes of Los Altos.
Evenings or weekends, through Outschool. Students join from home in small groups, never one-on-one.
You get the Studio, the teacher console and my ten-week teaching guide. Your teacher can follow the guide, adapt it, or teach it entirely their own way. I train them either way.
Drive times are approximate, from downtown Los Altos on a weekday afternoon.
The AI writes the code. Deciding what to build, writing the plan, testing what came back and deciding what to keep are the student's work, and that is what the course teaches. Lesson 3 is spent checking the AI's own claims.
Not to build. The Studio builds only while a class is running with an adult present. Students can open what they've made at home to show their family.
No. The teacher runs the room and the lessons. Training covers the console, the ten weeks of lessons, and what to do when something goes wrong.
A current web browser on any school device: Chromebooks, iPads, laptops, touch or keyboard. Your IT team adds a short list of addresses to the content filter, ideally a few weeks ahead.
It's kept only as long as the course needs it and then deleted on a fixed schedule. No student names are ever entered.
Anthropic's Claude models, run under Anthropic's requirements for organizations whose products serve minors.
The Studio runs in a browser, so there's nothing to install. Every student project is served from its own subdomain, which keeps one student's project from reading another's. The wildcard entry matters most and is the easiest to miss.
app.aibuilderlab.org api.aibuilderlab.org admin.aibuilderlab.org *.artifacts.aibuilderlab.org
A short call: I'll show you the Studio and the teacher console, and we'll work out which option fits your school.