Yes, you can build a small app today without any programming skills, as long as you work with an AI agent such as Claude Code or Codex. What is realistic is a clearly defined tool for a specific task. You still need to learn to read code roughly, interpret error messages and test your app.
This article shows where the line between achievable and wishful thinking lies, which skills you should pick up and what a first project can look like in practice.
What is realistic without prior experience
Vibe coding means developing software with the help of AI. You describe in plain language what the application should do. The AI agent writes the code, creates files and runs commands. You check the result and give the next instruction.
A lot is achievable for a first project, as long as the scope stays manageable. Typical building blocks you can implement well with an agent:
- display a list of entries, for example contacts
- a detail view that shows all the information about one entry
- validate input before it is saved
- change a status and filter by status
- export data as a file, such as a CSV
It gets harder when the app is supposed to manage a company's real customer data from day one, with user accounts, permissions and several connected systems. The agent can generate code for that too. Without experience, though, you can hardly judge whether that code is secure and does the right thing in every case. That judgement is exactly what you build up with a small project.
What you still need to learn
The AI agent does the typing for you. The judging stays with you. Four skills decide whether you end up with an app you can trust.
Breaking an idea into testable steps
"I need an app for my contacts" is not yet an instruction. An agent will turn it into something, and you won't know whether it is right. Small steps with a clear expectation work better. These are called acceptance cases. For example: "If I try to save a contact without a name, a message appears and the contact is not created." Or: "If I set the status to done, the contact disappears from the open follow-ups filter."
You can give sentences like these to the agent as instructions. Later you check exactly these sentences.
Reading code without writing it yourself
You don't need to be able to write every line yourself. You should, however, be able to find out where something happens in your project. Where is a status change saved? Which file builds the contact list? You can ask the agent what a function does and have it point you to the place in the code. Over time you will recognise patterns, even though the terms in the code are usually in English.
A short log helps: what did the AI contribute, and what did you change yourself? That way you can explain later why your app works the way it does.
Interpreting error messages
Error messages are not a disaster, they are clues. Many tool messages appear in English. Read the top line first and look for the file name and line number. Then give the agent the whole message and describe what you were doing at the time. "It doesn't work" helps it very little. "After clicking Save, this message appears and the contact is then missing from the list" helps a lot.
Testing instead of hoping
An app that worked once won't necessarily still work tomorrow. Tests are automatic checks you can repeat at any time. An example from a follow-up app: is a follow-up overdue? Check three cases, with a date of yesterday, today and tomorrow. Failure cases matter just as much: empty input, invalid email addresses, an interface that doesn't respond. And secrets such as access keys do not belong in code that others can see.
What you need to get started
Before you write your first instruction, your computer needs some basic equipment. This checklist helps you check:
- a laptop or desktop running Windows, macOS or Linux (a tablet is not enough)
- permission to install software, which is often restricted on company devices
- Node.js version 20 or later
- Git, to save changes in a traceable way
- your own access to Claude Code or Codex
- a willingness to look up English terms and messages
How to set up these tools step by step is described in the freely readable Mission 0 on setting up the tools (in German).
What a first project can look like
A good first project solves a small, familiar problem. A contact and follow-up app for a small business is a useful example: it has a list, a detail view, simple rules and a question that comes up often in everyday work, namely who still needs to be contacted.
The path could look like this:
- Clarify the brief: Who is the app for, and what should it do? Write it down in a few sentences before the agent generates the first line of code.
- Build the interface: A contact list with sample data and a detail view for each contact.
- Define saving and workflows: Validate input, add status changes (for example from new to contacted), filter for open follow-ups.
- Connect an interface: Practise on a simulated interface first. That way you see what your app does with an empty response, an error or a long wait.
- Find bugs and check access: Narrow down and fix bugs in a targeted way, and check who is allowed to see and change what.
- Explain and hand over: Write down how someone else can start and understand the app.
- Extend: Implement a new requirement, for example exporting overdue follow-ups as a CSV file, and test the new feature.
The order is no accident. Each step builds on a tested state. When you later work on an idea of your own, such as a list of open quotes or repair jobs, you can follow the same sequence.
Where the limits of a first project lie
A learning project is not a finished product. If your app stores its data in the browser, that is fine for practice data but not suitable for real company data. Things like secure data storage, backups or a well thought out management of access rights are missing.
A clean project therefore includes stating its limits openly. What can the app do, what can't it do, and what would have to happen before it is used in day-to-day operations? If you can answer these questions, you have understood more than someone who only shows off a running interface.
How the Vibe Coding Foundation course fits in
The Vibe Coding Foundation course from BUILT IN CLASS follows exactly the path described above. You build a contact and follow-up app with Claude Code or Codex, in eight missions from setup to CSV export. The course overview with all missions (in German) shows the goals and planned times in detail.
The course is aimed at working adults and small teams with no programming experience. You learn online and at your own pace, with no fixed dates and no videos to watch along. The planned time is around 30 hours in total, for example six weeks at about five hours per week. Each mission has a goal, criteria in the form of tests and hints at three levels in case you get stuck. On your computer, a command checks the criteria. What counts is the check on the academy's server in an isolated environment.
The four skills from this article are the competences the programme certifies: breaking tasks into testable steps, building an app with an agent and changing it in a traceable way, securing failure cases, data, access and secrets, and explaining and transferring your own solution. At the end there is a written online exam with six questions about your project. An AI prepares the assessment, and an examiner awards the points. How this works is described under exam procedure and rules (in German). Once you pass, you receive a private academy certificate, not a state qualification.
You can find costs and what is included under price and services. Note that your own access to Claude Code or Codex comes on top.