AI coding makes better software? Many developers see it differently and name the problems | News

There is no question that AI solutions can be helpful tools in many areas and increase productivity – especially when it comes to quickly collecting large amounts of data. However, whenever a factual representation is required, caution is required, because LLMs, as gifted storytellers, are not necessarily the most truthful people – especially since an LLM does not really understand what they are writing. An industry trend is now agentic solutions that take on multi-step tasks, as well as specialized tools for creating program code. In theory, this even enables users without programming knowledge to create apps. However, there is very mixed feedback from experienced developers, as 404 Media summarizes in an article.

AI changes are often little questioned
Several developers explain in a sentiment report that they are pressured to use AI tools, but the results often generate more rework, uncertainty and technical ballast. However, the accusation is in no way directed against AI as an occasional aid. Several of the people interviewed acknowledged that such tools can be very useful when prototyping, exploring unknown areas of code, summarizing logs, or finding documentation.

Often good for components, but not for projects
It becomes problematic when AI agents create large-scale changes to complex code bases and companies then focus primarily on quantity, speed and visible AI usage. Those surveyed are particularly critical of the effort required to check the results. For small auxiliary functions, suggestions can be checked comparatively easily. However, it looks different when you are faced with a huge mountain of foreign code: With major changes in branched projects, security-relevant components or historically grown code bases, things become significantly more difficult.

“It makes me stupider”
There is also a psychological effect, as several developers report that the constant use of AI makes them less confident in everyday tasks. Routines that were previously taken for granted suddenly have to be looked up again – or even questioned again. “It definitely makes me stupider,” is how a quoted voice puts it into words. Newbies in particular achieve quick results without recognizing any problematic side effects. This, in turn, shifts the responsibility even more to experienced, more expensive developers – which can negate cost advantages.

More differentiated consideration required
The question is therefore not whether AI can be helpful in software development, because it undoubtedly is, depending on the area of ​​application. More code generated does not automatically mean progress or more productivity – especially if you end up with less understanding of your own code base and architecture. The key here is to decide where automation makes sense, but without confusing “a lot of code created” with “much better developed”.