Building websites without programming knowledge is possible today in just a few hours with AI tools like Manus.You describe in natural language what should be created and receive a functional site.The term Vibe Coding has become established for this way of working: developing via prompts instead of source code, with the result as the benchmark and not the path to it.

How tempting this is and where it reaches its limits is shown by a practical example from prompt engineer Susanne Renate Schneider.The case revolves around an AI-created website that suddenly became unreachable.Precisely this moment is interesting — not the creation, but the operation afterwards.

Creation is the easy part

In web projects, the vast majority of the effort does not arise with the first version, but afterwards.A site must remain reachable, content must be updated, security vulnerabilities must be closed, data protection and accessibility requirements change.Anyone who regards the first version as the actual project systematically underestimates what follows.

AI generators optimise for the first impression.They deliver a visible result, and that is their strength.The questions that determine operation are usually not asked by them: Where is the site hosted?Who has access?What happens if the provider changes its offering?And who is responsible when something does not work?

Four points that do not appear in the prompt

  • Hosting and reachability: If the site is delivered on a platform of the AI provider, its availability depends on that provider's business model, tariffs and technical decisions.Your own domain and your own hosting contract are not a detail, but the foundation for a website belonging to you.
  • Maintainability: Generated code is functional, but not necessarily comprehensible.If nobody understands the structure, every small change becomes a new attempt — and in the worse case, a rebuild.
  • Handover capability: Can content be maintained by employees without anyone having to prompt?Without an editorial interface, a dependency on a single person arises.
  • Responsibility: Who is liable if a form processes data incorrectly or an imprint is missing?An AI tool does not assume this responsibility.It remains with the operator of the website.

What prompt-based prototypes are good for

This does not follow as an argument against AI-assisted creation.It follows as an argument for using it in the right place.Prompt-driven prototypes are extremely useful when a result needs to become visible quickly and does not yet have to last:

  • Making ideas tangible internally instead of discussing them in concept papers
  • Placing variants of structure, tone or page layout side by side and evaluating them
  • Landing pages for a single, time-limited campaign
  • Internal tools and dashboards with a small, known user group

What these cases have in common is that an outage is tolerable.Nobody loses revenue if a click dummy is unreachable.

When a stable platform becomes necessary

The situation is different as soon as a website is part of business operations.As soon as enquiries arrive through it, applications are accepted, products are to be found or legal obligations must be documented, the benchmark shifts.Then what counts is not how quickly the first version was there, but whether the site can still be changed in three years.

Proven foundations exist for this area.A content management system such as TYPO3 — that is, software with which editorial teams maintain content themselves — comes with permission management, multilingualism, versioning of content and a defined update path.This is not an end in itself, but the answer to precisely those questions that become visible in the Manus case: who may do what, where the data lies, and how to return to an earlier state.

A pragmatic path

In practice, both approaches can be combined.AI tools clarify in a short time how something should look and feel.This clarification is valuable and replaces long rounds of coordination.The sustainable implementation then arises on a platform that reflects operation, maintenance and responsibility.

It is important to make this decision consciously and not let time pressure replace it.A prototype that silently becomes the production website is the most common path into precisely the situation in which a site goes offline without warning.

What this means for web and software projects

AI shifts the costs in web projects to later: the entry becomes cheaper, operation does not.Anyone who uses vibe coding as a tool for concept phases and internal applications and chooses a maintainable platform with clear responsibilities for business-critical presences gains speed without giving up control.

The decisive question is therefore not whether AI can build the website, but who will bring it back online on Monday morning.

Sources