Three questions, three answers: AI – the productivity boost for coding?

AI-powered software development is shaking up the industry. While some rave about productivity boosts, others warn of mountains of technical debt. And even the previously critical Linus Torvalds is already experimenting with vibe coding in a side project. How should developers position themselves now? Christian Winkler, lead author of iX 2/2026, explains where the dev job becomes more productive with language models and where natural intelligence cannot be replaced after all.

Prof. Christian Winkler

is a data scientist and machine learning architect. He holds a PhD in theoretical physics and has been working in the field of big data and artificial intelligence for 20 years, with a particular focus on scalable systems and intelligent algorithms for mass text processing. He is professor at Nuremberg Institute of Technology since 2022. His research focuses on the optimization of user experience using modern methods. He is the founder of datanizing GmbH, a speaker at conferences and author of articles on machine learning and text analytics.

Vibe coding is now on everyone's lips: Is this really the future of software development?

Vibe coding is a very exciting application and definitely useful for quick prototypes at the moment. Therefore, it is definitely part of the future of software development. However, it seems unlikely to me that developers will become unemployed because of it. Quite the opposite could be the case: routine tasks will be implemented by AI in the future, and developers can focus on the really difficult problems. Fortunately, the mentioned routine tasks also include implementing tests. This is a rather unpopular task among developers, who should be pleased with AI support.

What can I use as a guide for model selection? How helpful are benchmarks?

There are several benchmarks, such as SWE-bench or the Coding Benchmark from LMArena. However, there is a risk that the models are too optimized for the benchmarks. Therefore, if possible, one should try out the models themselves or orient oneself towards benchmarks that are updated very often or are reviewed with individual requirements by humans.

With the US flagship models from the cloud, there is also the risk that you will pass on your sensitive code as training material. What alternatives do companies have?

Models hosted within the EU should not have this issue because the legal jurisdiction is different here. So, one could limit oneself to providers who keep their data in the EU. Some US providers promise exactly that. An alternative would be to run models yourself whose weights have been published. With open models, all requirements for data protection and data security can be met.

Christian, thank you for the answers! More details on how to code more productively with AI are available in the new iX. We also examine whether the vibe coding platform can keep its big promises and provide an overview of the legal issues, such as liability, when using AI agents. All this and many other topics can be found by readers in the February issue, which is now available in the heise Shop or at newsstands.

In the series “Three Questions and Answers,” iX aims to get to the heart of today's IT challenges – whether it's the user's view from the PC, the manager's perspective, or an administrator's daily routine. Do you have suggestions from your daily practice or that of your users? Whose tips on which topic would you like to read briefly and concisely? Then feel free to write to us or leave a comment in the forum.

(axk)