Stay calm. Keep going. AI must wait or the bubble will burst.

13 Aug, 2025

We are once again living in exciting times.

I feel that one of the many controversial topics is coming to a head, and I'm talking about AI, or more specifically, the LLM hype. Since the launch of ChatGPT in November 2022, things have been going uphill in every respect, and the question is, aren't all the discussions surrounding AI culminating in some kind of vague form of... panic?

I mean, the extremes have increased: On the one hand, there are the enthusiasts who swear by the incredible benefits of AI, who see a tenfold increase in productivity, who know AI agents that can take on almost any task in the field of software development, and who anticipate ever more powerful LLM models, which are then promptly rolled out in ever shorter cycles and with ever greater fanfare.

And on the other side, there are those who are, shall we say, disappointed. The added value of LLM is not recognized there; on the contrary, there are reports of negative effects on productivity, security gaps, or other types of frustration. This quickly mixes with the unfortunate behavior of the big AI corporations, which act more like rabbits on speed than reputable, trustworthy partners and do not paint a good picture of the prevailing digital capitalism.

My observation is that not only have the extremes of experience increased, but AI has also gained increased focus and advocates, especially in the business environment, as a new miracle cure. Of course, the FOMO¹ effect is to be expected here; no company wants to be left behind, and if everyone is promising such incredible success and benefits from AI, then you have to be part of it and jump on the bandwagon. But a major problem is and remains that, as things stand today, these are still just promises.

When software benchmarks talk about >70% achievement, that still means <30% of tasks or goals are not being adequately accomplished. And especially in software, a discipline so closely related to mathematics, whose beauty lies in the fact that correctness—100% correctness—is achievable and is the premise on which everything is built, are values < 100% celebrated as success? Who even points out that the benchmarks often only cover popular programming languages such as Python, NodeJS, or even Rust²? Don't our industry CEOs know that a large part of their assets are based on languages such as C/C++, Fortran, and Cobol, that the real “golden nuggets” are secret to most companies and are often fed from extreme technical niches with equally specialized solutions, i.e., programming languages or models?

It is what it is, but the odds are clearly in favor of the pro-AI proponents right now. Nevertheless, there are growing voices that see parallels to the dot-com crash of the 2000s. Personally, I also believe that we need massive consolidation and more common sense, but that we will get it. Of course, this will be accompanied by considerable financial turmoil, which is what a crash like this entails.

If I am wrong, and no one can predict the probability today, it would indeed mean a massive shift for the future technological world. This would have noticeable consequences for the entire industry and also for me personally, which I and everyone else affected would have to deal with.

However, I am a “late adopter” and think that this philosophy helps me, especially in this unclear situation. The technologies I work with today and that drive my industry and many others are those that have been around for a long time and have proven themselves over time³. The best example is SQL, or rather relational databases/models, which have now been around for 46 years. The UNIX philosophy and GNU/Linux as a “spin-off” in the world of operating systems – GNU software was first developed in 1984, meaning it has been running smoothly here for 41 years, not to mention the proprietary UNIX precursors. What else? Von Neumann architecture, Internet Protocol Suite, various programming paradigms – all of which have been around for a while.

If the technology currently understood as AI lives up to its promises, then it would certainly be mentioned in the same breath as the above examples. And it will remain. Generations of engineers and technicians will grow up with it, learn it, and be able to tame it. Not always for the best for humanity, but that's another topic and can be proven and discussed based on experience with many other technologies, especially those outside of information technology.

The conclusion? The soup is being cooked very hot right now, so I'm waiting until I taste it. To all those who are as confused as I am and still think critically, I appeal to you: don't let yourselves be carried away and driven crazy by the loud-mouthed snake oil salesmen; wait until the patients either leave unhealed or actually come out of the tent healed by the oil. In the meantime, take care of your responsibilities, be good people, spend as much time as possible with your family and friends and in nature, and as little time as necessary with AI bubblegum. Stay calm and carry on.

Translated with help from DeepL.com (original in German language)