In software development, day-to-day work has shifted noticeably in a short space of time. AI assistants – that is, tools that suggest code, explain it or generate entire sections – now take over routine tasks that used to be typically assigned to newcomers. Some companies conclude from this that junior positions can be cut.

Golo Roden, CTO and a well-known voice in the German-speaking developer community on architecture and team issues, considers precisely that a serious mistake. In essence, he describes doing without junior hires as the least sensible decision companies can make in the current situation. His reasoning: precisely because AI is fundamentally changing the work, companies must not neglect the next generation.

The argument behind the provocation

The reasoning is less socially romantic than it appears at first glance. Senior developers are not created by hiring, but by years of project practice. Anyone who cuts the entry level interrupts the pipeline from which the experienced staff of the day after tomorrow come. The effect occurs with a delay – and is then difficult and expensive to correct.

There is also a quality aspect. AI-generated code often looks plausible, but is not automatically correct, secure or maintainable. Someone has to check the suggestions, place them in the context of the existing architecture and, if in doubt, discard them. This ability to judge comes from experience with real systems, not from operating an assistant.

What shifts in the entry phase

The fact that junior roles are needed does not mean they stay unchanged. Realistically, the focus shifts:

  • Less time spent manually writing standard code, more time spent reading, understanding and evaluating existing solutions.
  • More engagement with tests, failure patterns and debugging – that is, with the question of why something does not work.
  • Earlier contact with business requirements, because translating customer wishes into technology remains the actual core work.

Anyone who lets newcomers simply wave through AI output builds no know-how, but dependency. The learning effect arises where a suggestion is questioned, traced and either accepted or rejected with justification.

Why this concerns clients

For companies that commission web or software projects, the service provider's team structure is not an internal detail. It determines whether a system can still be maintained in three or five years – and by whom.

A team consisting exclusively of a few very experienced people is vulnerable: if a key person drops out or leaves, there is no substructure that could take over knowledge. A team without experience, in turn, may deliver quickly, but without robust quality assurance. The viable variant lies in between: mixed seniority with clear review rules.

How to recognise good practice

A few questions help assess a development partner's way of working without delving deep into technical details:

  1. How is the team composed, and who checks whose work?
  2. Are there binding code reviews, i.e. the review of changes by a second person before they go live?
  3. How are AI tools handled – are there rules on which data may be entered and who is responsible for the results?
  4. How is knowledge documented so that it does not depend on individual heads?
  5. Are newcomers systematically supported, or do they just run alongside?

The answers say more about a project's reliability than the list of tools used.

AI does not replace responsibility

AI assistants noticeably reduce the effort for recurring tasks. But they only shift the place where quality arises: away from typing, towards evaluation. Responsibility for architecture, security, data protection and maintainability remains with people – regardless of who or what delivered the first draft.

That is precisely why the entry level is not a cost-saving item, but an investment in the review competence of the coming years. Anyone who cuts it saves personnel costs today and pays later with missing knowledge in their own environment or at their service providers.

Assessment for web and software projects

For ongoing projects, this means: do not just ask whether a partner uses AI, but how they safeguard the results and who is responsible for them. A balanced team with functioning review processes is a better indicator of maintainability than any tool list. And where the next generation is missing, the experience needed to assess AI output at all will also be missing in the medium term.

Sources