Over the past few weeks, I have spent a good part of my time going back over my Kubernetes course. Not just fixing a few pages or adding commands. I wanted to review the whole thing: the explanations, the progression, the exercises and, above all, the way we check that a skill has really been acquired.
At the same time, another topic keeps coming up in my LinkedIn conversations with IT directors, CTOs, CISOs and executives: what should we do with the applications that business teams are starting to build with artificial intelligence?
These two projects may look far apart. To me, they answer the same question: how do we go from something that works to something we know how to understand, secure and run?
Kubernetes: I wanted to go much further
My Kubernetes course has existed for quite some time. It has grown with each version, with feedback and with the needs of certification preparation.
But as content kept piling up, I needed to step back.
So I ran a complete audit. Some pages were corrected, others enriched, the certification paths were reworked and new exercises were added.
Today, the course amounts to 14 modules, 136 lessons and about 170 hours of work, counting reading, hands-on practice and assessments. It relies on 1,895 quiz questions.
And, on the practical side, the site catalogue lists 129 Kubernetes labs. The course builds 123 of them directly into its lessons, including 3 capstones and 3 mock exams.
In my opinion, the course has become very complete. It covers first deployments as well as administration, networking, storage, security and cluster troubleshooting.
But the figures are not what I find most interesting.
From guided labs to real troubleshooting situations
For a long time, I offered exercises to apply what had just been explained. That is essential when you are learning.
But following a procedure and solving a problem are two different skills.
Take one of the new challenges.
You arrive on a Monday morning. Several people changed a cluster over the
weekend. The API server no longer responds, a worker is NotReady, one
application restarts in a loop and another does not start.
This time, nobody tells you which commands to run.
You have a situation, symptoms and results to achieve. It is up to you to observe the cluster, read the logs, identify the causes and fix it.
What interests me is not that you know a kubectl command by heart. It is that
you know where to look when things do not go as planned.
That is why I added more challenges, and also capstones: broader exercises that require you to combine several skills with no step-by-step guide.
DSOXLab also changes the way you practise
All of these exercises rely on DSOXLab, the open source tool I am developing to prepare the environments and check the results.
The Kubernetes labs run on real virtual machines, with real clusters. We do not simply simulate command output in a browser.
DSOXLab prepares the environment, presents the mission and then checks the state you actually obtained.
Must a Pod be available? The test checks that it is. Must a service respond? It has to respond. Must a node be repaired? It has to be operational again.
Grading does not depend on the commands you used, but on the result.
For me, this is an important change. I do not only want to offer exercises that can be completed. I want learners to be able to measure what they can really do.
CKA, CKAD, CKS: going all the way to mock exams
Certification preparation has also evolved a lot.
The CKA, CKAD and CKS paths now have capstones and hands-on mock exams, with tasks to carry out, a time limit and an assessment based on the state of the cluster.
The aim is to come close to the working conditions of the exams, without claiming to reproduce the official tasks.
The KCNA and KCSA preparations are also there, with their content and their knowledge assessments.
Finally, the new CKNE (Certified Kubernetes Network Engineer) certification, announced by the CNCF in November 2026, has joined the landscape. It covers cluster networking: CNI plugin, Services, DNS, Gateway API, network policies and traffic observability.
I have started to bring these topics into the course, but CKNE preparation is not finished yet. Some objectives are already covered, while others still need content and exercises.
I would rather say so than present a certification as fully prepared when it is not.
The list of Kubernetes labs is available for those who want to start practising.
A new topic is making itself felt: AI-generated applications
While I was working on Kubernetes, another line of thought was taking up more and more space.
For the past few weeks, in the comments and conversations triggered by my LinkedIn posts, the same question keeps coming up among IT managers and executives.
Business teams now use AI tools to create applications themselves. In a few hours, they can present a form, a dashboard or a small management tool.
I fully understand the appeal. Needs that had been waiting finally get a first answer.
But what happens next?
Who checks the code? Who controls the dependencies? Who decides whether the application can access company data? And above all, who will take responsibility for it when its users come to depend on it?
I had already raised this question in my post “Your time saved must not become our technical debt”.
The conversations that followed convinced me that I needed to go further and build a proper course.
Securing and governing AI-generated applications
That is the title of this new project.
I do not want to create yet another course that only explains how to use Cursor, Claude Code or another assistant to produce an application.
The subject is different.
I would like to show how an organisation can welcome these applications, assess their risks, qualify them and possibly integrate them into its information system.
The idea is to start from a concrete case: the purchasing portal, a tool that a buyer created with the help of AI and that she now wants to connect to the orders of an ERP.
Rather than answering yes or no every time, we are going to build the technical means to make an informed decision.
I also want to keep two points of view: that of the decision makers, who must accept or refuse a risk, and that of the technical teams, who must put the corresponding controls in place.
The first module, “Understanding AI-generated applications in the enterprise”, is published, in French and in English, in the course on AI-generated applications. Its 7 lessons cover vibe coding, shadow AI, risks, the hosting architecture, the classification of applications from C1 to C4, the legal framework and governance. Each lesson has its quiz. Three DSOXLab labs let you practise: classifying applications, reviewing an agent's changes against a specification, and checking the dependencies it proposes.
What comes next is under design. The programme I am preparing will cover, in particular, the classification of applications, the responsibilities shared between business teams and the IT department, the protection of code and dependencies, CI/CD controls, environment isolation, access authorisations to the information system and operating conditions. It will rely on an isolated qualification and hosting platform, with a forge, a CI, signed artefacts, a Kubernetes zone, an API gateway and go-live decisions made by humans. I would rather build something genuinely usable than announce a finished course too early.
Learning to do, but also to own what you build
In going back over Kubernetes, I wanted learners not to stop at a successful deployment. They also need to know how to diagnose, repair, secure and maintain.
With AI-generated applications, I find exactly the same concern.
Producing an application is no longer necessarily the hardest part.
Understanding it, securing it and keeping it maintained remain very real responsibilities.
This is the direction in which I want to keep developing the site.
As always, the courses, the exercises and the teaching tools remain free, with no advertising and no mandatory subscription.
If this work helps you make progress, you can also support the project on Ko-fi, once or on a regular basis. It helps fund the time and the resources needed to keep everything going.
And above all, I would like to hear from you: what is still missing from the Kubernetes course? And if you are an IT director, CTO, CISO or head of a technical team, how do you currently manage the applications created with AI outside the usual development channels?