DecompileD 2025

Besides the very prominent topic of AI, this year we were especially looking forward to the presentations featuring best practices from various fields. In terms of coding, we enjoyed the presentation by Andreas Roth, who demonstrated using an example of a frontend technology how "Don't repeat yourself" (DRY) as a development mantra can also lead to excessively rigid and complex code. Therefore, it is important to work with abstraction layers as well. Since these can also have disadvantages, one must always weigh the options. Andreas presented a simple, triangular graphic of reusability, interface complexity, and scope, into which one can categorize their components. We definitely want to try out this method.

As a service provider, we must always be able to understand our customers and their needs. In his presentation, Martin Betz emphasized that it is not only about fulfilling customer wishes. One should question whether each feature really contributes to the core functionality of an application and whether there are other ways to achieve it. It is mainly about empathizing with customers and advising them to the best of our ability, especially when they struggle to clearly articulate their wishes. Although all of this is nothing new, we found it important to see these things compactly in front of us. Even though Martin's presentation was more geared towards large companies developing their own products for end customers, we can still take away some key points from his approach.

It should not have escaped anyone in our industry that AI currently often dominates discussions on the topic of development. Therefore, it is clear that it also plays a major role at a conference like DecompileD. We listened to two presentations on this topic.

Yishai Beeri discussed the impact of AI on DevOps workflows and how they will evolve to keep up with rapid advancements. Yishai stated that the trend will continue, with code being significantly more generated by automated agents in the future, and there will be a need for processes to ensure that humans still review it in a meaningful way. It is also important to determine where only the result of an automated test is relevant in the end, and at which point the second or even third look of a human is necessary.

Stefan Queißer spoke about the difference between myth and reality regarding the question of whether AI will one day replace developers. The answer has already been expressed in many places: No, but the human element could potentially evolve to a higher level than just writing code in the future. Particularly interesting was Stefan's view that Artificial Intelligence only provides added value when it performs its tasks invisibly in the background, with humans receiving a designated interface for it. This implies that one may not necessarily need to know that an LLM or something similar is running. In short: As long as one needs to review and correct the results of AI, it cannot replace human work.