Your kid ships a real app in two weeks.
An online school where kids learn AI by using it for real. They team up with an AI agent on their own cloud computer, build their own idea, and launch it to real people. No experience needed.
Est. first class · Summer 2027Why learn this now.
This is Precedence Research's 2026 valuation ($900B) compounding at their forecast rate of 18.7% a year, computed for this exact second — about $5,300 more every second you spend reading this, on track to cross $1 trillion this August.
Precedence Research, 2026 → 2035 forecast ($4.2T)jobs asking for AI skills are growing eight times faster than the job market overall
PwC Global AI Jobs Barometer, 2026average wage premium for workers with AI skills
PwC Global AI Jobs Barometer, 2026One login. Everything they need.
No setup on your family computer. Signing up gives them their own computer in the cloud — a full desktop with a browser, a real terminal, and an AI agent ready to build. Five stages, two weeks, one launched app.
And they're never alone in there: a buddy lives on their screen the whole time. It narrates every move the agent makes in plain language, points at exactly what to look at, and when something's worth understanding, it pauses the whole build — the AI literally can't take its next step — until your kid gets it. Then the world un-freezes and the building continues. Teaching, built into the machine.
-
Learn
Drive the agent, don't just watch it.
First, the basics that make everything else work: what context is, why agents forget, and how to ask for what you actually want. Short quizzes with unlimited retries — kids drill until it feels easy.
-
Practice
Real terminal, training wheels on.
Guided reps on real examples: moving around a terminal, reading what the agent did and why it did it, plus the first basics of telling people about what you make.
-
The idea
Pick something worth two weeks.
They talk the idea out with their agent until it's really theirs, then shape a plan the way builders do: landing page first, make it great, then the app itself, one phase at a time.
-
Build
Their own computer in the cloud.
They get a real machine with Halden ready in the terminal. They approve the plan, then watch it build while we narrate what's happening in plain language. Tips arrive at the right moment — like using image models to mock up the perfect UI before building it.
Live, while the agent worksTOOL · edit_file → dogwalks/index.htmlIt's fixing your headline before your colors. Structure first, paint second — that's on purpose.
-
Launch
Real users, or it didn't happen.
Their agent helps them get it out there: where to share it, what to say, who it's for. Then they watch real people use the thing they made. And a real product can charge real money — if it's good enough, their app can be earning before the two weeks are up.
0 days, start to shipped.
What they walk away with.
The launched app is the trophy. These are the skills underneath it — the ones that transfer to whatever they build next.
- Agent literacy
- How AI agents actually work: what context is, why agents forget, how to plan with one, prompt it well, and steer it when it drifts. The mental model, not just a tool — so it survives whatever tool comes next.
- Terminal confidence
- Moving around a real computer with a real command line: running commands, reading output, understanding what the machine is doing. The same environment professionals work in.
- Building in phases
- How real products get made: landing page first, make it great, then the UI, then the features. Scoping an idea down to something a two-week sprint can actually finish.
- An eye for design
- Using image models to mock up a UI before building it, judging what looks right, and making it better. Taste is a skill; they practice it.
- Marketing basics
- Naming the audience, writing the pitch, picking where to share it, and getting their first real users. The part most builders skip — theirs won't.
- A shipped product
- A live URL they made themselves, real people using it, and the story of how. The start of a portfolio, and proof they can finish things.
Not another coding class.
AI camps and coding classes exist, and some are good. Here's where Halden splits off.
-
Security, built in.
Everything runs on our cloud computers — the AI, the accounts, the keys, the whole build. Nothing gets installed on your machine and there's nothing sensitive for a kid to hold or leak. They get the full power of the real tools, with zero danger to them or your computer.
-
The agent is narrated, live.
Others hand kids an off-the-shelf coding tool and an instructor. Halden watches the agent work alongside your kid and explains every move in plain language, with tips arriving at the right moment. That's built into the product, not a slideshow.
-
Mental models before tools.
Most programs teach a specific tool. We teach how agents think — context, memory, steering — before the build starts. Tools change every year; the mental model doesn't.
-
Parents get a real window.
Progress, projects, and what the agent did, visible to you in plain language — not a certificate PDF at the end. Rare in this space; standard at Halden.
For parents.
You can watch the whole thing. Parents get a clear window into their kid's progress: what they've learned, what they're building, and what the agent did for them, all in plain language.
And the AI stays in our hands, not theirs. We manage the accounts and the keys with sensible limits built in, so kids never touch a credential. The pilot cohort is deliberately tiny — a real person watches every session and adjusts the course around your kid.
Full disclosure: this page is the first thing Halden shipped. The survey behind that button is the second.
Questions parents ask.
What ages is this for?
The sweet spot is 11–15. Younger or older kids are considered case by case during the pilot, since every session has a real person in the loop.
Does my kid need coding experience?
No. If they can type and hold a conversation, they can start. The first stage teaches them how to drive an AI agent before anything gets built.
What equipment do we need?
Any computer with a web browser. The building happens on a cloud computer we provide — nothing gets installed on your family machine.
How much time does it take?
About one to two focused hours a day, for two weeks. The phases are designed so each day ends with visible progress.
Is the AI safe for my kid to use?
The AI accounts and keys stay in our hands — kids never handle a credential. Sessions are supervised, every agent action is visible in plain language, and you get a window into their progress the whole way.
What does it cost?
Pilot pricing isn't final yet — honestly. The survey behind the Join button asks what feels fair, and those answers set the number.
What do they actually launch?
A real app at a real URL, shared with real people — family, friends, their school, a community that fits the idea. Their agent helps them market it in the final phase. And it's a real product: if it's worth paying for, it can take payments and make actual money.
What if they get stuck?
The buddy on screen catches most stumbles the moment they happen — and behind it, the pilot cohort is deliberately tiny, so a real person is watching every session and steps in before frustration does.
Is this only for kids?
Kids first — that's who nobody else serves safely. But the same buddy, cloud computer, and two-week arc work at any age, and plenty of adults want to learn to build with agents too. If that's you, take the survey and say so — adult cohorts are on the roadmap.
When does the first class start?
Estimated summer 2027, with a handful of pilot students before that. Survey folks hear first, and the 5%-off form locks your spot at the front of the line.
The first cohort is small. Want in?
Join nowTwo minutes, and your answers decide what we build next. First class estimated summer 2027 — founding spots go first.