With a project, you learn to code because you build something yourself and can see from clear criteria whether it works. Watching, on the other hand, often leaves you feeling you have understood everything, until you sit in front of your own screen and have no idea where to start.
This post explains why a project in missions with testable criteria works differently from a video course, how staged hints keep you moving without doing the thinking for you, and what a starter kit is.
Why you get stuck with video courses
In a video, someone else makes every decision. They know which file comes next, which command works and why an error appears. You see the finished approach, but rarely the detours that led to it.
On top of that, there are a few typical stumbling blocks:
- The pace is set for you. You rewind, pause, type along. You are busy copying rather than understanding.
- Your screen looks different. A different version, a different error message, often in English. That is exactly the point where the video does not stop. It just keeps going.
- Nobody tells you whether your result is right. If your page looks roughly like the one in the video, you still do not know whether it actually does the right thing.
With an AI agent such as Claude Code or Codex, the problem shifts. The agent writes code quickly. Your job is then to give it a clear brief and judge whether the result fits. That judgement comes from trying things out, checking and correcting them yourself. Watching barely builds it.
What a project in missions does differently
A project in missions means you work on the same application from start to finish, and each mission builds on the one before. In the Vibe Coding Foundation course, that application is a contact and follow-up app for a small company, filled with sample data.
The route looks roughly like this. First you describe who the app is for and what it should do. Then you build a contact list and a detail view. After that you validate inputs and add status changes and filters. Later you connect the app to a simulated interface and practise what happens when a response is empty, an error comes back or the response takes its time.
The difference from a video: if your filter for overdue follow-ups does not work, you have to find out why. That is exactly where the learning happens.
One goal and one planned time per mission
Every mission has a clearly stated goal and a planned time. The planned time helps you judge whether you are on track or whether it is time to get help. It is a guide, not a race.
Criteria instead of gut feeling
Each mission comes with criteria in the form of tests. On your own computer, a command checks whether your app meets them. That gives you feedback that does not depend on your mood.
An acceptance case might read like this: if you try to save a contact without a name, a message appears and the contact is not created. That can be checked. Either it happens or it does not.
What counts in the end is the check on the academy's server. It runs in an isolated environment with no network access. That makes it clear the result comes from your project and not from a setting that just happens to work on your machine.
How hints in three stages help
When you are stuck, it is tempting to look straight at the finished solution. That feels good but achieves little, because you have skipped the key idea. So every mission has hints in three stages.
The idea is that you open only as much help as you need right now. If the first stage is enough to get you moving again, the rest stays your own work. If it does not help, you go one stage further. Along the way you learn exactly where the gap in your understanding was.
This fits well with working alongside an AI agent. There too, it pays to narrow a question down yourself before asking the agent to rewrite everything. An example: your status change from "open" to "done" is forgotten after a reload. Instead of asking the agent to "fix everything", you ask specifically where the status is stored. And then you actually understand the answer.
In the course you also keep an AI log. In it you note what the AI contributed and what you changed yourself. It sounds like extra work, but it shows where you made decisions and where you simply accepted what you were given.
What is a starter kit?
A starter kit is a prepared starting project. It contains a working foundation, so you do not begin with an empty folder and spend hours hunting for settings. You start the app, see that it runs and build from there.
This matters most at the beginning. Many people give up before they write any code, during setup: missing programs, wrong versions, unclear error messages. A starter kit lowers that hurdle because the framework is already in place.
In the course, Mission 0 takes care of this. You prepare your computer, your AI access and the starter project, and launch the app for the first time. For that you need a laptop or desktop running Windows, macOS or Linux, permission to install programs, Node.js version 20 or later, and Git. You can see exactly how it works in Mission 0: setting up your tools, which is free to read (in German).
Worth knowing: the learning project stores data in the browser. It is meant for practice and is not a product for real company data.
Checklist: are you learning or just watching?
You can use these questions to check any way of learning, including the one you are using now:
- Are you building your own project that will be able to do something specific at the end?
- Before each step, do you know how you will tell that it worked?
- Do you get feedback on whether your result meets the criteria?
- Can you get help in measured doses instead of seeing the whole solution straight away?
- Can you explain what the AI wrote and what you changed yourself?
- Can you extend your project with a new feature that nobody demonstrated first?
If you hesitate on several of these, it is rarely down to you. Usually the learning format is missing the part where you have to act yourself and check your own result.
How the Vibe Coding Foundation fits this approach
The Vibe Coding Foundation course is built exactly this way: online, at your own pace, with no fixed dates and no videos to watch along with. It is aimed at working adults and small teams with no programming experience. You work with Claude Code or Codex through your own access, and its costs come on top of the course price.
The planned time is around 30 hours in total, of which around two hours are for the exam. One possible pace is six weeks at about five hours a week. You will find all missions with their goals and planned times in the course overview (in German), and the process step by step under how the course works.
Mission 7, the transfer mission, is particularly revealing. There you extend your app with a data export: overdue follow-ups are exported as a CSV file, and you test the new feature yourself. This is where it shows whether you can apply what you have learned to a new requirement.
Along the way you collect badges, XP and levels. They do not replace the criteria, but they make your progress visible. You can see which ones exist on the badges and levels page (in German).
The course ends with a written project exam, entirely online. If you pass, you receive a private academy certificate from BUILT IN CLASS, not a state qualification. What the course price includes is set out under price and what is included.