And a question to start with: why does an airline pilot need to undergo periodic simulator training?

Recently, I worked on a fairly large brownfield consolidation project: 350 GPU servers in two weeks… it was intense. I wanted to take it on because it felt like a return to basics for me, although this time I used a well-known model. What actually prompted me to write, though, was that doctors are facing a similar situation regarding the different impact of AI on seniors and juniors (https://www.nytimes.com/2026/09/25/opinion/ai-doctor-medical-students.html?unlocked_article_code=1.EFE.775t.hdxPUF46-TX6&smid=nytcore-ios-share). So, I feel the need to draw a few conclusions, deliberately leaving out the technical details.

For seniors

AI has become a reliable tool, requiring only a few incremental instructions, as long as you keep it within your thinking process:

• helping to absorb different concepts and syntaxes, with references to manuals and guides
• correlating data and identifying gaps – work that used to be very tedious and time-consuming
• precise, example-based automation: batches of four hours of work done in two minutes?
• cleaning up lint (small inconsistencies in configuration files, some with potential consequences)
• a critical second opinion (things I had not thought of or was not aware of, at just the right moment) – the general prompt helped here too

Not so good (yet):

• catastrophic design errors (missing connections between leaf switches in a backend without spine switches, as if you had a city with boulevards between neighborhoods but no roads between houses)
• missing subtle details that I know from experience
• troubleshooting loops that follow no logical thread, but rather focus on what has not been tried yet
• running out of credit at a critical moment (during the migration window) – oops, luckily I had a personal card in the system, but I had not seen that coming

But with a model that keeps getting better, the biggest mistake you can make in the long run is to delegate understanding. To stop discussing things with colleagues to get a second opinion. Over time, I have understood many things and discovered design errors simply by explaining something to someone else. On the other hand, it seems incredible: one extra competent colleague for 40 euros over two weeks. Yes, that cheap.

For juniors

The model helped me because I could use it directly (I no longer needed to know Python or delegate the work). I could check the result myself, knowing what I wanted. But two years ago, that would not have been the case: I would have delegated to DevOps (because I do not know Python that well). Back then, just as now, a junior DevOps engineer joining the team would have been one more person to coordinate and check under pressure. But that is not the point I wanted to make: for a junior DevOps engineer, the very person who worked hard to learn Python, this was a missed opportunity to put that knowledge into practice and make mistakes. In other words, the opportunity to develop instincts and cognitive associations through raw mistakes (“learning the hard way”), but also through someone else’s varied experience, has suddenly shrunk. What can you do as a junior? Keep learning something thoroughly, know how to verify it, but also find out where things are happening and make sure you are there (in other words, ask to be involved). I will come back to this idea.

Cognitive and social skills (connecting things, understanding and systematizing, anticipating, organizing, explaining, presenting, selling an idea) are no longer developed in the same way. But we still need them just as much, so they have to be exercised, like going to the gym. As I said above, you need to want to create opportunities for yourself and understand what you are doing. That is the only way to go beyond a bar that keeps rising, and that makes the difference between colleagues who will all be using the same basic set of tools and knowledge.

Structure and understanding

Not to mention writing concisely and following a logical thread (engineers hated writing even before, and now they would be even less inclined to do it). Sure, AI can write documentation, but volume and perfectly written details do not amount to a logical thread (say, the order in which you build out the infrastructure). For the person reading it (another human being), it is not as engaging, and the writing will certainly have massive gaps in structure and content – things the reader will never manage to piece together.

Why is structure important, and why does it need to be practiced and articulated rather than delegated? As an anecdote, I witnessed the chaos that ensued when a vendor used AI to answer questions asked with AI (by engineers who were not sure about the subject). There was a clear mental fog: not only was there no logical thread, but people no longer even knew who had asked the question or what information was missing. Nothing clear and a massive waste of time, in other words. So, do not give up expressing yourself and thinking logically. Personally, if I provide a logical structure and bullet points to address or complete, the outcome is entirely different.

For managers

If seniors no longer delegate, who will train the next generation of seniors? This is fundamentally a selection problem, not merely an individual one. Opportunities must be created. Juniors will quickly learn to recognize the places where they can learn and grow.

“De-skilling”, “never-skilling”, “re-skilling”

Generally speaking, we are not talking about “de-skilling”, which does not have such a major impact on seniors, but about “never-skilling” (see the article linked above). The issue is not avoiding AI, because productivity and precision matter, but the lack of critical thinking that comes from training juniors with AI and, ultimately, the loss of control. Even though AI models make our lives easier, they must be used strictly as tools (“re-skilling”), not as a substitute for experience:

• never delegate understanding or the logical thread, never ever; do not lose your curiosity, which is something AI does not have; this applies to both seniors and juniors
• do not stop expressing yourself to other people; it is something that must always be practiced, even if someone else can do it for you; especially for juniors
• do things in small pieces and put them together until you understand the process and the whole; if you are a junior, do things manually and check them with AI at first, not the other way around
• create opportunities for yourself and find your own path, one that preserves your unique set of skills: after all, people have always found a way forward after a shock.

So, why are airline pilots required to undergo periodic simulator training? Exactly, so they do not forget how to fly the plane manually. And why do we need a pilot instead of a robot? It is a matter of trust that things will not go off the rails. But the pilot did not choose to go to the simulator on their own either…