Digital products ·
What I mean by vibe coding, and where the magic of a prompt ends
In plain language, vibe coding is a way to create digital products through a conversation with artificial intelligence: describe an intention in everyday words, receive working code and immediately test the result. It greatly shortens the distance between an idea and a prototype. Yet as a prototype moves closer to a real product, the work depends less on magic and more on architecture, verification and responsibility.
The term and its original meaning
Andrej Karpathy introduced the term “vibe coding” in February 2025. He described an almost carefree mode of working: state a request, accept the proposed changes, see what happens and continue the conversation with the model without examining every line of code.
This approach is excellent for a small experiment, a personal tool or the first proof of an idea. It can turn something that might have remained a note or mock-up into a working artefact. But it is imprecise to call every form of AI-assisted development vibe coding. Once a product stores data, takes payments or becomes part of a business, intuition alone is no longer enough.
A prompt does not replace the Concept
A good prompt may produce a convincing screen, but it does not answer the central questions: who the product is for, what problem it solves, what should happen first and which constraints cannot be broken. Without those answers, speed merely helps build an arbitrary form more quickly.
That is why I begin by articulating the Concept rather than asking for code. I define the person and context, construct the scenario, and divide the system into data, functions and states. The AI then receives a direction in which every decision can be tested, rather than a vague wish.
Speed demands verification
Code that works on the first attempt is not yet a product. Edge cases, security and data privacy, interface accessibility, third-party service costs, performance and long-term maintainability all need to be examined before release.
Researchers studying vibe coding also point to the risks of opaque codebases and gaps in responsibility. For me, the practical rule is simple: AI may propose and execute, but a person must understand and own the decision. The greater the cost of failure, the stricter the verification.
How I work
I came to digital products through architecture, object design and running my own production workshop. In each of those fields, an idea passes through material, construction, budget and real use. Vibe coding changes the material of the work, but it does not remove that path.
Within the Method, I use AI as an execution environment: I build a prototype quickly, test the scenario, identify weak points and progressively turn a Concept into a working website, SaaS product or internal tool. The value lies less in the amount of code produced than in carrying the original intention into a form that people can use and continue to develop.
From an idea to a digital product
If the task calls for a service or internal tool, we can define the first complete journey and a path to a working release.