In February 2025, AI researcher Andrej Karpathy wrote on X about a new way of programming. He described what he wanted to an AI, accepted every change without reading it, and simply pasted error messages back into the chat. He called it "vibe coding." By the end of the year, the Collins Dictionary named it the word of the year for 2025.
Today, marketing teams, sales managers, and CEOs are building their own small tools this way. Some of them save weeks of work. Others expose customer data. In this article, we’ll clarify what vibe coding means, what companies can use it for effectively, and where the limits lie.
What is vibe coding?
Vibe coding is a form of software development where you describe what a program should do to an AI in plain language. The AI writes the entire code. You only check if the result works and provide feedback. You barely read the code itself. Andrej Karpathy coined the term in February 2025.
Karpathy is a co-founder of OpenAI and was formerly the head of AI at Tesla. His post was read millions of times because he put a name to something many developers were already trying out themselves. One detail often gets lost: Karpathy himself wrote that the method works quite well for weekend throwaway projects. There was no mention of software for enterprise use.
Vibe coding: What does it mean?
"Vibe" refers to a feeling or atmosphere. Loosely translated, vibe coding means programming by feel. There isn't a standard German term that has caught on; even here, everyone just calls it vibe coding.
The name hits the nail on the head: you work with the result you see. In vibe coding, no one checks whether the underlying code is clean or secure. This distinguishes it from the AI-assisted programming used by development teams today. Developer Simon Willison pointed this out early on. According to Willison, if you read, test, and understand every line an AI writes, you are using it as a fast writing assistant. Vibe coding begins where no one looks at the code anymore.
How vibe coding works in practice
A typical process looks like this:
- Describe the idea: You write what you need into a tool like Lovable or Claude. For example: "Build me a calculator that calculates a quote based on a daily rate, project days, and travel expenses, and exports it as a PDF."
- View the first version: After a few minutes, a clickable application is ready in your browser.
- Test and refine: You click through the app and describe what is missing or not working. If an error pops up, you copy the message into the chat.
- Repeat: This loop continues until the tool does exactly what it’s supposed to do.
You don't need any programming skills for this. The ability to describe requirements clearly is helpful. If you've ever written a good brief for an agency, you have the most important skill.
Want to learn step-by-step how to build your own applications without coding experience? Check out our detailed guide on coding with AI.
Vibe Coding Tools: How to get started
The main difference between these tools is how much of the technical side you actually see. Browser-based tools like Lovable hide the code almost entirely, while editors like Cursor show it to you line by line.
For your very first attempt, ChatGPT or Claude in your browser will do just fine. Both can build small applications directly within the chat and display them right there.
Examples: How companies are using Vibe Coding
The best use cases have one thing in common: they are small, clearly defined, and won't cause any damage if something doesn't work quite right.
- Clickable prototypes: Planning a customer portal or an app? You can have a draft ready in two hours to show your management team or early customers. It makes your agency briefing much more concrete.
- Internal calculators: Quote calculators, margin tools, or ROI calculators for sales teams can replace those Excel files that only one person understands.
- Campaign landing pages: You can create a page for an event or webinar without having to submit a ticket to your web agency.
- Data processing: A small script can take product data from various supplier lists and convert it into a consistent format.
- Mini-dashboards: Turn an export of your sales figures into an overview that shows exactly the metrics you need every week.
Get a free strategy call and discover how AI can help you automate workflows, create better content, and scale your marketing.
Book a call
Limitations and risks of Vibe Coding for businesses
What doesn't matter in a weekend project can quickly become expensive in a corporate setting. Two cases from 2025 show where the problems lie.
Security: In the spring of 2025, security researchers examined 1,645 apps built with Lovable. 170 of them had exposed databases, allowing outsiders to access email addresses, payment data, and API keys. The vulnerability is documented as CVE-2025-48757. The database access rules were missing, and users didn't notice because the apps were running perfectly fine.
Control: In July 2025, SaaStr founder Jason Lemkin built an application using Replit's AI agent. On the ninth day, the agent deleted the production database containing data on over 1,200 executives, even though Lemkin had explicitly issued a freeze on changes. The agent then claimed the data could not be recovered. Lemkin managed to restore it using the rollback function anyway. Replit subsequently separated the development and production databases.
Maintainability: Code that no one understands cannot be maintained. This often only becomes apparent months later. An interface changes, an update breaks something, and the person who built the tool is on vacation. Beyond a certain size, the AI also starts going in circles: it fixes one bug while creating two new ones.
Data protection: Many vibe-coding tools run on US servers. If you copy real customer data into a prompt or load it into an app you've built yourself, you need a solid foundation that complies with the GDPR. You can read about what you should look out for with US tools in AI and data protection.
Shadow IT: When employees build their own tools, IT often knows nothing about it. No one knows the access credentials, and no one is making backups. It’s the same pattern as with Shadow AI, just with self-built software.
The biggest catch: vibe-coding apps look finished long before they actually are. An interface that works well reveals nothing about open databases, missing backups, or access credentials hidden in the code.
When is vibe coding worth it for a company?
Whether vibe coding is a good fit depends on the use case. A simple classification helps:
- Good fit: Prototypes, drafts for presentations, calculators without personal data, and landing pages without forms. If a tool like this fails, you just build it again.
- Only with review: Internal tools that run permanently or work with company data. Here, someone with development experience should check the code and access rights before deployment.
- Hands off: Anything involving customer data, payments, or access to systems like ERP and CRM. In these cases, vibe coding is at most a way to create a prototype. The final version should be built by a development team.
If you want to allow vibe coding within your team, we would start with four rules:
- Use test data: Keep real customer and employee data out of the mix until someone has vetted the tool.
- Separate environments: AI agents shouldn't have access to production systems. The Replit incident shows why.
- Review before going live: Anything that runs permanently or is customer-facing needs to be checked by someone with development experience first.
- Keep an overview: A simple list of tools, their purpose, who's responsible, and access levels is all you need.
Vibe coding is not a substitute for software architecture: "Rapid prototypes bring a massive boost in innovation to your team. However, to prevent speed from turning into security gaps or shadow IT, clear guardrails for access rights and data privacy are essential right from the start." — Jens Bohse, Co-Founder
Bottom line: Vibe coding turns ideas into reality in hours
Vibe coding is all about telling the AI what you want and testing the result without having to read the code. It’s a powerful tool for prototypes, internal calculators, and quick drafts. Ideas that used to sit in the backlog for weeks can be clickable by the end of the afternoon. But as soon as customer data, payments, or production systems are involved, you need rules and someone who understands the code. The cases of Lovable and Replit show what can happen otherwise.
The best way to get started: Pick a small internal problem that doesn't involve personal data, like a quote calculator, and build it in an afternoon using Lovable or Claude. After that, you'll know what the method can do and where its limits are.
If you want to figure out where Vibe Coding fits into your company and what rules you need for it, get in touch with us for an AI consultation.
FAQ: Frequently asked questions about Vibe Coding
Is Vibe Coding secure?
Only to a limited extent. AI-generated code often contains security vulnerabilities, such as exposed databases or credentials left in the source code. For prototypes without sensitive data, the risk is manageable. Anything that handles customer data or is publicly accessible should be reviewed by someone with development experience before going live.
Why is it called Vibe Coding?
The name comes from Andrej Karpathy. In February 2025, he described how he gets completely lost in the feeling of programming and forgets that the code even exists. "Vibe" refers to the mood or feeling of the process.
Do I need programming skills for Vibe Coding?
No. Tools like Lovable or Replit work entirely through natural language. However, having some basic knowledge helps when things don't go as planned or when you need to assess whether an app is secure.
Is Vibe Coding good or bad?
It depends on how you use it. For prototypes and small internal tools, Vibe Coding saves a lot of time and money. For software that needs to run long-term and protect data, it isn't enough on its own.
Can you make money with Vibe Coding?
Yes, some founders have built and sold their first products this way. For companies, the biggest benefit is usually time saved, such as through faster prototyping and fewer feedback loops with agencies. If you want to sell a product, you have to take security and maintenance very seriously.
What is the difference between Vibe Coding and No-Code?
With Vibe Coding, the AI writes actual code that you can export and continue to develop. No-code tools like n8n or Make work with pre-built blocks that you connect with clicks. This makes no-code more limited, but also more predictable.
Posted by:
Jens is the co-founder of bakedwith, a boutique consultancy firm specialising in smart automation and AI. He helps medium-sized companies and corporations to optimise processes, reduce manual work and achieve growth through efficient workflows. Previously, he was Growth Lead at OMR and worked as a freelancer, helping numerous companies optimise their CRM and automation systems. Jens is passionate about combining growth and efficiency to enable teams to focus on what matters most.
.webp)
.webp)
.webp)
.webp)
%201.png)
