Sketching out a dashboard, adding functions step by step and sharing the result with the team via a link: this is exactly the approach described in a current practical report on Claude Code, Anthropic's development tool.The point here is less about the technology than about the target group.Anyone working this way no longer needs programming skills, but instead formulates in plain language what the tool should do.

The term vibe coding has become established for this way of working – it refers to prompt-driven development in which an AI model writes the code and the person in front of it mainly describes, checks and readjusts.The source code is still created, it is just no longer typed by hand.

Why this resonates in business departments

In many mid-sized companies there is a long tail of small requirements that never get a project budget: an overview of open quotations, an evaluation sheet for production planning, a form that replaces three Excel files.Such requests either end up in the IT queue or in yet another spreadsheet.

Prompt-driven tools shift this boundary.A marketing manager can see a first version of her campaign dashboard within an hour instead of describing it in a requirements document.The value here lies less in the finished tool than in the clarity: a running prototype answers questions that a concept paper leaves open.

The difference between prototype and production system

But this is also exactly where the misunderstanding begins.A tool that works for three colleagues is not yet a system that will hold up in an organisation over years.Between the two lie several issues that are easily lost in the rapid creation process.

Data protection and access rights. As soon as a self-built dashboard displays real customer, HR or revenue data, the same rules apply as for any other application.Who sees which data?Where is it stored?Is the link via which the tool is shared actually used only internally?A tool distributed via a link without login is unproblematic for a demo, but as a rule not for real data.

Maintainability. Code that nobody in-house has read is difficult to develop further.As long as the original person continues working with the same AI tool, this works well.If they leave the company or an interface changes, the department is left with a tool that nobody can explain.This is not an argument against vibe coding, but one for a deliberate decision about which tools are handed over and which deliberately remain disposable.

Integration. Most of these prototypes initially live in isolation.But the benefit only emerges when they connect to existing systems – to the CRM, the ERP or a TYPO3 backend in which content and editorial processes are maintained anyway.This connection is the part that requires experience with interfaces, authentication and data models.

Guidelines for use in the company

A pragmatic framework prevents many small successes from turning into an opaque shadow IT:

  • Distinguish two categories: Experiments with test data may be created freely.Anything that processes real data or is used by more than a handful of people goes through a review.
  • Test data first: Build prototypes with anonymised or invented data sets.The question of whether the tool is any good can be answered just as well this way.
  • Keep a register: Who built what, which data sources are connected, who is the contact person?A simple list is enough.
  • Define a handover point: From when does IT or an agency take over?Sensible triggers are real data, external use or a business process becoming dependent on the tool.
  • Deliberately discard prototypes: The insight gained remains even if the code is rewritten.This is often cheaper than subsequently hardening a grown makeshift solution.

What changes for collaboration

For service providers, the role shifts.If business departments bring their ideas to a working draft themselves, a project no longer starts from scratch but from a concrete template.Requirements workshops become shorter because less text and more screen is discussed.What is needed instead are more strongly the skills that a prototype does not provide: data modelling, permission and role concepts, operations, accessibility and the connection to existing systems.

What remains realistic: prompt-driven development lowers the barrier to entry, not the requirements for a production system.The work shifts from writing the code to assessing, securing and embedding it.

For web and software projects, this means concretely: it is worth explicitly allowing business departments to prototype while at the same time drawing a clear line as to when a tool belongs in orderly channels.Those who define this line early gain speed in the idea phase without having to deal with a collection of unmaintainable isolated solutions in two years' time.

The prototype is then what it should be – a very good requirements specification.

Sources