Tutor
Hints and explanations, never the answer. Replies in your language.
Build real projects with AI at your side, and keep the hard parts that make skills stick.
For a long time, lectures were the centre of a course. Now a clear explanation of almost anything is one question away, so we moved the centre to building. Each course is a path of quests. A quest is one real thing to build, with a clear definition of done.
Hints and explanations, never the answer. Replies in your language.
Use coding agents freely. The tests and the defense still check that you understand it.
Build it yourself, with hints only. For the parts worth doing by hand.
Ask the group. AI joins a thread only when someone asks.
People who check the close calls and notice when you're stuck.
Every quest comes with a short, hand-picked list of lectures, articles, papers and docs. Each one says how long it takes and why it is worth your time, so you read and watch with a purpose.
Every day brings a new model, a new tool, a new paper. It is easy to read about all of it and build nothing. A quest gives you one thing to do next, the few sources that matter for it, and a finish line.
You learn by building something that has to work. Along the way you will get stuck, and some of it will fail. That is the part that teaches you. At every step you are also questioned: by tests, by a tutor, by an examiner and by people.
It knows your quest and your latest failing tests. It explains and gives hints, never the solution, and replies in your language.
Your code runs in a sandbox against cases you haven't seen. You find out right away whether it really works, and when it doesn't, you fix it and run again.
It reads your code, your write-up and how you worked with your coding agent, then scores a short rubric and says what to improve.
An AI examiner asks about your code and choices. An agent can write code for you, but only you can explain it.
My calendar shows the wrong week when I pass a Sunday.
What does weekday() return for a Sunday? Try it in the shell, then look at how you find the Monday.
Fixed the date check. Running again…
All 4 suites pass
To improve: show your agent the failing test before asking for a fix.
A member opens ?week= with a Sunday. Which days does your page show, and which line of your code decides that?
Monday to that Sunday. Line 18, where I subtract weekday() from the date.
Mastered
Mentors check the close calls. Borderline results and unsure grades go to a person, and every correction makes the AI graders better.
It comes back later. After you master a quest, short review questions return over the following weeks, so what you learned stays.
You learn two things side by side: how to build software with AI coding agents, and how to build AI features into software. It ends with your own AI app, live on the internet.
Decide what you'll build and ship at the end of the course.
about 4 hoursBuild a small app with an agent, then add a real feature to an app someone else wrote.
about 6 hoursGive the app an AI assistant that answers from its data and documents, then open it to Claude and Codex.
about 8 hoursBuild the app you planned, ship it on the internet, and show it.
about 16 hoursPeople who review the close calls and notice when you're stuck.
Ask the group. AI joins a thread only when someone asks it to.
People on the same quests, a step ahead of you or a step behind.
Moments like this don't come often. Tools that once needed big teams and big budgets now run on a laptop, and they get better every month.
One person with an idea and the skills to build it can now make something real: for their studies, their community, or a business of their own.
It is still early, and there is room for everyone who learns to build.
Learn to build real things, and understand how they work.
Ship AI products that work, and know why they work.
Pick up the new way of working without losing the old skills.
Start building