Work With IA After 8 month/en

Six mois avec l’IA : entre passion, frustration et puissance décuplée

When I started using AI intensively, I had a rather strange feeling: I became fascinated by this imaginary friend capable of doing some things better than me, faster than me, but not always with as much perspective.

At first, it is exciting. Then, very quickly, it becomes frustrating.

Frustrating because AI moves fast, very fast. It suggests, it generates, it corrects, it documents. But it does not always understand the real context, the hidden constraints, the history of a system, or the small subtleties that can make a theoretically correct solution dangerous in production.

After about six months of working with AI every day, I understood something important: for now, we are complementary.

I really mean for now, because I am not sure this partnership will remain the same in the future. When we look at the work of people like Yann LeCun, who are trying to give machines a deeper form of reasoning, it almost feels like the story has not even really started yet. And yet, everything is already moving very fast.

AI IN MY WORK AS A CLOUD AND INFRASTRUCTURE DBA

In my work, I am a DBA and I mainly work in the cloud, with still a part of on-premise infrastructure. My role constantly requires me to move from one topic to another: databases, networking, operating systems, application design, security, automation, DevOps pipelines, scripting, monitoring, and troubleshooting.

A modern DBA can no longer limit themselves to writing SQL. They need to understand the complete ecosystem around the database.

With the massive rise of automation, this role is expanding even further. We need to work with new languages, sometimes return to older ones, understand cloud tools, industrialize deployments, make processes more reliable, and maintain consistency across increasingly complex environments.

Personally, I have become very comfortable with PowerShell to automate tasks in Azure. I also use Python when needed. Today, I create my own applications, modules, and tools to manage several thousand production databases.

In my case, we are talking about more than 4,000 databases.

With Azure Batch, Azure DevOps, well-designed pipelines, and a solid automation strategy, we can achieve almost surgical precision. We can orchestrate processes, manage exports, automate backups, standardize deployments, monitor errors, intelligently retry certain processes, and produce usable reports.

But all of this requires one essential thing: discipline.

VERSIONING HAS BECOME VITAL

One point I now consider fundamental is versioning.

When Microsoft, Oracle, Amazon, or others change something, even slightly, it can put the stability of a system at risk if everything is not properly controlled.

A module version, a package version, a script version, a pipeline version: everything must be clear, traceable, and reproducible.

Without versioning, we end up chasing problems. With versioning, we can understand, compare, roll back, and secure changes.

This is even more true when we start working with AI, because it can produce a lot of code very quickly. If we do not structure what it generates, we can quickly end up with chaos that is very well presented.

THE IMPORTANCE OF INSTRUCTION FILES AND AGENTS

What truly changed my experience with AI was not just asking questions to a chatbot. It was learning how to structure the context.

I use instruction files that define the working context for my AI sessions:

```mediawiki

• naming conventions;
• file structure;
• error handling rules;
• commenting standards;
• internal best practices;
• company logic;
• tags;
• pipelines;
• modules;
• project architecture.

```

I also started working with a logic of specialized agents.

For example:

• a PowerShell-oriented agent;
• a documentation agent;
• a SQL expert;
• an MSSQL expert;
• an Oracle expert;
• an agent for exports;
• an agent for project structure;
• an agent for troubleshooting.

On top of that, I use a kind of orchestrator agent that chooses the right expert depending on the topic.

This is where the experience becomes truly powerful. We are no longer simply asking AI: “write me a script.” We give it a framework, rules, history, a way of thinking, mistakes to avoid, and standards to follow.

And at that point, the quality changes completely.

ERROR FILES: A VERY VALUABLE TECHNICAL MEMORY

One of the most interesting aspects of my approach is error management.

When AI makes a mistake, I do not only want to correct the final result. I also want to prevent it from making the same mistake again in the next generation.

For this, I use Markdown files dedicated to known errors, traps already encountered, and corrections to apply.

These files become a kind of technical memory. They allow AI to know about past mistakes and better respect my constraints in the future.

This is particularly useful in automation scripts, where a small PowerShell scope issue, a forgotten parameter, a naming mistake, a wrong path, or an incorrect retry logic can have significant consequences.

With this method, AI does not become perfect, but it becomes much more useful. It improves with my context.

THE FEELING OF MANAGING A SMALL TEAM

Today, when everything is well structured, I sometimes feel like I am managing a team of several dozen people.

Not in the sense that AI truly replaces a human team, but in the sense that I can quickly call on several “profiles”:

• a PowerShell developer;
• a technical writer;
• a SQL analyst;
• a reviewer;
• a documentation specialist;
• a DevOps assistant;
• a troubleshooting expert.

It is quite magical.

But this magic only works if I remain in control.

AI greatly accelerates things, but it needs a pilot. It can generate quickly, but it does not always know what is acceptable in a production context. It can propose an elegant solution, but not necessarily one that fits the history of an infrastructure. It can give a convincing answer, but one that may be false or incomplete.

That is where human experience remains essential.

AN EXPERIENCE THAT HAS BECOME POSITIVE

Après six mois d’apprentissage intensif, de tests, de frustrations et d’ajustements, mon expérience avec l’IA est devenue très positive.

Je ne la vois plus comme un simple outil de génération de texte ou de code. Je la vois comme un accélérateur de réflexion, de documentation, d’automatisation et de structuration.

Elle m’aide à aller plus vite, à explorer plus d’options, à mieux documenter, à standardiser mes pratiques et à produire des outils plus solides.

Mais elle m’oblige aussi à être plus rigoureux :

• plus clair dans mes instructions ;
• plus structuré dans mes projets ;
• plus attentif au versioning ;
• plus précis dans mes règles ;
• plus critique dans mes validations.

Finalement, l’IA ne remplace pas mon métier. Elle augmente fortement ma capacité à l’exercer.

ET DEMAIN ?

Je reste positif, mais aussi prudent.

L’évolution est tellement rapide qu’il est difficile de prévoir où nous serons dans deux ou trois ans. Si les machines commencent réellement à raisonner de manière plus profonde, la relation entre l’humain et l’IA pourrait changer très vite.

Aujourd’hui, je me sens complémentaire avec l’IA.

Demain, je ne sais pas.

Mais une chose est sûre : l’histoire ne fait que commencer.

Et pour ceux qui acceptent d’apprendre à travailler avec ces outils, pas seulement les utiliser superficiellement, le potentiel est énorme.

Pour un DBA, un ingénieur infrastructure, un développeur ou toute personne qui doit naviguer entre plusieurs technologies, l’IA est déjà un outil remarquable.

Pas parfait. Pas magique sans méthode. Pas sans risque.

Mais bien utilisée, avec du contexte, des règles, du versioning et une vraie supervision humaine, elle devient un levier impressionnant.

Après six mois, je peux le dire clairement : malgré les frustrations du début, je ne reviendrais pas en arrière. </syntaxhighlight>