At Anthropic, the company behind the AI assistant Claude, a manager has formulated an assessment that is causing discussion in the industry: the classic coding interview is dead.The reasoning given by Felix Rieseberg is uncomfortably simple.Small, self-contained programming tasks are now solved better by their own model than by most applicants.Anyone who tests precisely this ability in a job interview is therefore measuring something that is already automated anyway.
According to this reading, the whiteboard task should be replaced by prototypes: things actually built and functional, from which it can be seen how someone thinks.Rieseberg names taste, empathy and principle-guided thinking as criteria – that is, the ability to make good decisions, adopt user perspectives and work according to comprehensible principles.Add to that the formula "Hunger beats tenure": self-drive counts for more than years of service.
A shift, not a free pass
For newcomers, this initially sounds like empowerment.Anyone who builds something convincing without a long professional track record can qualify themselves.At the same time, however, the bar shifts significantly: instead of a clearly defined technical examination, entrepreneurial thinking is expected.This requirement is harder to grasp, harder to train and presupposes a kind of constant commitment that not every life situation allows.
It is worth reading this classification soberly.The statement comes from a company that produces AI assistants – a certain self-interest in the thesis that models will take over programming is obvious.Nevertheless, it is still easy to observe in practice that self-contained tasks are now reliably handled by assistant systems.
What is actually scarce in projects
The actually interesting point for clients lies elsewhere.If writing individual functions becomes cheaper and faster, the bottleneck in software projects shifts upwards – to the levels that an AI does not clarify on its own:
- Architecture: How are systems cut so that they remain maintainable in three years' time?Which dependencies are deliberately entered into, which are not?
- Requirements clarification: What is the software actually supposed to do – and what of that is merely a wish from a single meeting?
- Quality assurance: Generated code often looks plausible.Whether it is correct, secure and performant only becomes apparent through tests, reviews and operation.
- Responsibility: Who is liable for ensuring that a solution meets data protection requirements and also works under load?
These tasks have always been demanding.What is new is that they make up the predominant part of value creation when implementation itself loses weight.
Prototypes as a means of communication
The fact that prototypes are gaining importance in the application process has a counterpart in project work.A functional draft answers questions that can be discussed for weeks in concept papers.How does the operation feel?Does the chosen approach hold up for the real data set?Do business departments understand the process without explanation?
With AI support, such drafts are available faster than before.This changes the project logic: instead of specifying for a long time and then building once, something can be shown earlier and corrected on the basis of real feedback.What remains important is the separation between prototype and product.A draft may take shortcuts; a system in productive operation may not.
Consequences for collaboration with service providers
Anyone commissioning web projects should adjust their evaluation standards accordingly.The question "How many developer hours are in this?" becomes weaker as a quality indicator.Other points are more meaningful:
- How does a provider justify its architecture decisions – and does it also name their disadvantages?
- What quality assurance is planned: automated tests, code reviews, security checks?
- How are AI assistants used, and who checks their results?
- What happens after launch?Maintenance, updates and further development are the longer part of the life cycle.
- Can the resulting code later be taken over by others, or does a dependency arise?
A note of caution: speed alone is not a quality feature.An application that is up in a few days can nevertheless become expensive over the years if the data model and interfaces are not thought through.Especially with systems such as TYPO3, which are operated and extended over long periods, conceptual care pays off more than speed in the first week.
What this means for teams
For developer teams, this follows less as a threat than as a shift in focus.What is needed are people who can assess an assistant's suggestions instead of merely adopting them – and who understand which business problem lies behind a requirement.That presupposes experience that cannot simply be replaced by rapid prototyping.Both belong together: the willingness to try things out and the judgement to assess the result.
What remains open is how junior staff will build this judgement in future when the classic entry-level tasks are increasingly handled by models.The industry currently has no convincing answer to that.
For web and software projects, this means concretely: the value of an agency lies less and less in the pure writing of code and more and more in architecture, quality assurance and the targeted use of AI assistants.
Anyone comparing quotes should therefore ask about decision logic, test coverage and maintainability rather than about effort line items alone. And anyone building teams internally should create room for prototypes – but equally for the experienced eyes that turn a prototype into a robust system.