Vibecoding makes founding easier than ever. Programmer Ole Zierau reveals his most important tips for getting started.
Artificial intelligence (AI) is driving the startup boom. The startup association counted more than 3,000 start-ups in Germany – in the first half of 2026 alone. “There has never been this much start-up momentum in Germany,” says association head Verena Pausder. The largest sector is the software industry with more than 800 start-ups. “AI significantly lowers the hurdles for starting a business and more and more people are taking advantage of this opportunity.”
Many founders build new apps and websites using vibecoding. Vibe coding describes a way of programming in which you simply explain to an AI what you want to build instead of writing the code yourself line by line. “The fascinating thing is that you no longer have to understand every single technical step yourself,” says developer Ole Zierau. “The models have knowledge from countless code bases, algorithms and technical documentation. This opens up completely new possibilities.”
Zierau is a principal engineer at the media and tech group Axel Springer. At our startup summer camp “Start-up scene is looking for super founders” he gets eight founders ready for vibecoding. For two months, the fellows build scalable consumer apps from scratch. Top professionals from tech and business will help you with this. At Gründerszene, expert Zierau shares his most important tips for getting started with vibecoding.
1. Ask the AI for vibecoding tips
With AI, your startup visions are just a prompt away, says Zierau. “The real change is that we no longer have to think about how to implement something technically. Instead, we focus on what we want to achieve. That How The model is increasingly taking over.”
When you get started, you still need “some basic knowledge,” says Zierau. “Just like founders also need a basic understanding of business relationships.”
Founders should understand: “How do computer systems work? How do applications work? What is a database and why do I need it? What is a backend, what is a frontend and what are the limitations of these systems?”
Zierau’s tip for beginners: Ask these questions directly to the AI model. “This gradually creates a question-and-answer dialogue that leads you to the desired goal.” And he advises: watch YouTube videos of developers streaming their work. “You learn an incredible amount.”
2. Find the vibecoding model that suits you
The real challenge is “understanding where you want to go and what technology you need to get there.” So describe your idea to the AI. Then ask how you can implement them technologically. “In my opinion, this is the best way to get started,” says Zierau. If you first want to build a prototype or try out an idea quickly, “tools like Lovable are the right place to start. You get to a result very quickly,” says Zierau. “So the barrier to entry is very low.”
Read too
For the “Superfounder” workshop, Zierau has prepared a technical basis on which the fellows can scale their apps. A Github project contains the base code for the user interface and backend. This processes data and controls the application. With the open source tool Terraform, the entire infrastructure can be rebuilt automatically and identically at any time. The AWS cloud platform also offers enough scope to significantly expand the applications later – for example with your own backend services or AI functions using tools such as Codex.
“The key is to have access to a model and then set up a development environment,” says Zierau. You can also ask the AI for guidelines for this. “This creates the basis for direct development. From this point on, the model opens up almost unlimited possibilities.”
3. Vibecoding AI keeps getting better
The AI can also make mistakes and hallucinate – like any chatbot. Zierau sees two solutions for this: “Either you discover an error yourself, copy-and-paste the error message into the model and let the AI analyze it. Or the model regularly checks the status of the application itself and looks for problems independently.” Monitoring tools that you can integrate directly into your vibecoding structure help here. They show you errors in the code without you having to learn a programming language.
Expect mistakes, say goodbye to perfection. “I would rather release an application that has not yet been tested down to the last detail in order to get early feedback and learn from it, rather than wait two years for the perfect release,” says Zierau. “In the end, it’s about trying out ideas as quickly as possible, learning from them and continually improving.”
Read too
Also trusts developers to continually improve their AI models: “The problems we see today will largely be solved.” Bugs that crop up today could improve AI in a few months. “We accept that we are not yet able to achieve 100% freedom from errors. But we are gaining enormously in development speed and can correct errors very quickly. We are currently seeing this trend everywhere,” says Zierau.
The quality will not always reach the level that experienced specialists used to achieve when building a database, for example. “But the models allow us to go from an idea to a working solution much more quickly.”
Zierau says about new software ideas: “It’s just one prompt away.” “You describe what you want to build and use the model to find the way there.”
The summer camp “Gründerszene is looking for the super founders” is a ten-week fellowship run by Gründerszene. Eight selected fellows are developing their own apps using artificial intelligence. The program is supported by our partners Dash0, DHL, OpenAI & Vercel.
