From repairing a Windows PC to organising files and analysing a phone, this first-hand account shows how AI and Codex can accelerate remote IT maintenance without replacing proven methods, backups or manual oversight.
How can artificial intelligence be used remotely to optimise a computer or phone?
It is a question I am exploring increasingly in practical terms.
I have no idea what young people who have been taught about artificial intelligence directly through their studies would think of my current level.
I originally learnt computing more than twenty years ago, although I had already acquired some basic knowledge as a child, around the time when the Internet became publicly available.
Today, I am therefore trying to understand the new vocabulary, the new technologies and, above all, the extremely rapid development of artificial intelligence.
In my view, there are two ways of looking at it.
Artificial Intelligence Is a Revolution… but Not Everything Is New
The first way of looking at the situation is to say that artificial intelligence represents a technological revolution.
And it does.
But there is a second way of looking at it.
For a long time, including in science fiction such as Star Trek, some of the principles we use today had already been imagined.
More practically, there are sometimes computing tools that are twenty or forty years old and can perform a task better than certain highly advanced commands produced by artificial intelligence today.
Not everyone is aware of that.
Even the best AI models will not necessarily rediscover spontaneously the reliable old computing technique that has worked for twenty years.
This is particularly true when dealing with:
- computer maintenance;
- security;
- workstation performance;
- system administration;
- diagnostics;
- repairs.
AI offers enormous possibilities, but it does not automatically replace everything that came before it.
Why Computing Is Particularly Well Suited to AI
One thing nevertheless strikes me: artificial intelligence can be particularly effective in computing.
That ultimately seems fairly logical.
Computing is a field that is:
- standardised;
- structured;
- documented;
- based on precise languages and commands.
What impresses me today are the demonstrations I see almost every day.
I watch commands scroll across the screen that I do not necessarily understand immediately in every detail, but I can appreciate the level of experience that would have been required to achieve the same result ten or twenty years ago.
At that time, professionals could spend years:
- studying;
- reading documentation;
- experimenting;
- specialising in a particular field;
- learning the right sequences of commands.
Today, artificial intelligence can produce some of those sequences in a few seconds.
Having studied computing myself and now working as a software developer in my company, I still find it particularly interesting to see demonstrations of this kind every day.
But you still need to know what to ask artificial intelligence and how to use what it produces.
Codex: Moving from a Prompt to Computer Action
In my work, I use Codex, among other tools.
It may be useful to explain simply what I mean by that.
Codex can turn instructions expressed as prompts into actions, commands or computer code.
When we use an artificial-intelligence web application, we usually see only the interface.
Behind that interface, however, there may be a great many components:
- different specialised AI services;
- APIs;
- paid processing services;
- traditional computer code;
- automations;
- deterministic sequences that sometimes have nothing intelligent about them in the modern sense of the word.
It is precisely this combination that interests me.
ChatGPT Can Produce Code; Codex Can Act Directly on the Machine
ChatGPT can already produce computer code.
For example, you can ask it:
Create an HTML and CSS page with the features I need.
JavaScript can also be added when the environment being used allows it.
In my use, however, Codex goes further: it can work directly inside the relevant computing environment.
That environment may be:
- a PC;
- a Mac;
- a Linux system;
- a server;
- and now, in my experiments, a phone.
We are gradually moving from:
“explain how to do it”
to:
“study the situation and carry out the appropriate operations in a controlled environment”.
For me, that is where things become genuinely interesting.
The Example of a Website or Sales Funnel
Let us take a relatively simple example.
Services such as Systeme.io already make it possible to create pages and sales funnels quickly.
ChatGPT can also be used to speed up that creation process.
Artificial intelligence can contribute at several levels:
- writing text;
- copywriting;
- page structure;
- generating illustrations;
- HTML;
- CSS;
- JavaScript when it is available;
- customising certain features.
This can make it possible to build things very quickly.
But every system also has its constraints.
If you use an existing platform, you benefit from its integrated features, but you do not necessarily have complete control over the whole system.
Conversely, when you build your own solution, you gain freedom but must also take responsibility for more technical constraints.
There are therefore always advantages, disadvantages and limitations.
Ultimately, that is true of almost every technology.
We Do Not Really Start from Nothing with Artificial Intelligence
One idea seems important to me.
We sometimes hear that artificial intelligence allows us to do everything immediately.
That is not quite how I see it.
We do not start from nothing with artificial intelligence.
We need to start with:
- an idea;
- a need;
- research;
- a minimum amount of study;
- preparation.
Once that work has been done, artificial intelligence can considerably accelerate implementation.
But it remains necessary to understand what we are trying to achieve.
That also explains why technical skills continue to matter.
Organising Information Is Already a Computing Task
Another project I am currently working on concerns my own data.
Over time, I have accumulated thousands of files.
Like many people, some of that information is:
- poorly organised;
- scattered;
- difficult to find;
- stored in different folders;
- sometimes forgotten.
I am therefore working on a document database indexed locally on my PC with artificial intelligence.
The aim is gradually to be able to:
- sort files;
- classify them;
- study them;
- retrieve information;
- understand what they contain;
- identify what I can produce from this mass of documents.
That is ultimately a good definition of computing work.
Computing is not only about computers and lines of code.
At its core, it is also about information and how it is structured.
A Much More Practical Case: Catching Up on Several Years of Windows Maintenance
I also recently worked on a Windows computer that was particularly difficult to restore.
The machine had accumulated approximately two to three years of overdue maintenance.
The installed version of Windows was no longer supported, and the computer had several problems.
In particular, it had been used with certain software or services obtained from unofficial sources.
That is a serious problem.
Pirated Software: The Risk Goes Far Beyond Price
I understand that people face different financial situations and that the history of the Internet is also connected with sharing content.
When I was a student, for example, we had internal networks that enabled students to share music, films and other files.
The context was very different from today.
Many professional software packages also cost several hundred euros, whereas numerous services are now offered through more affordable subscriptions.
But when a professional computer uses software from questionable sources today, the issue is not merely legal or financial.
It also becomes a technical and security issue.
Installing modified software or downloading files from untrustworthy sources can be an excellent way to compromise a computer.
For somebody who depends on that machine for work, taking such a risk to save the cost of a software licence does not generally seem logical to me.
Computer Security Has Also Changed Considerably
The security standards used today are no longer the same as they were fifteen or twenty years ago.
We now need to think at several levels:
- passwords;
- authentication;
- security codes;
- authentication applications;
- security keys;
- updates;
- system protection;
- account security.
The available options also depend on the hardware and operating system.
An older computer running Windows 10, for example, does not necessarily offer the same possibilities as a recent computer designed for Windows 11.
Security is therefore not simply a matter of installing antivirus software.
It forms part of a wider whole.
The Problem Was Not Just Windows
In the case of this computer, other constraints made the work more complicated.
The Internet connection was poor.
Access depended partly on tethering through a phone to avoid the cost of a conventional fixed broadband service.
I can understand that reasoning perfectly, because at one point I also wondered whether it was still useful to keep a fixed Internet connection when mobile plans offered so much data.
In practice, however, I noticed an important difference.
When working remotely on a computer for a long time, a stable wired connection remains far more comfortable.
Even in a large city, a mobile connection can experience:
- fluctuations;
- disconnections;
- lower speeds;
- network changes.
Wi-Fi itself can also be interrupted.
When remote maintenance lasts several hours, these small problems quickly become significant.
The person’s availability also had to be taken into account:
- when can I carry out the work?
- when can I connect?
- will the computer remain switched on?
- will the connection remain stable?
- how long might an operation take?
Remote computer maintenance is therefore not merely a question of technical commands.
A System That Was Almost Impossible to Repair Normally
The system was in particularly poor condition.
Some updates no longer worked.
System files appeared to be missing.
Even ordinary functions, including playing videos or using certain web services correctly, caused problems.
We needed to understand how Windows had reached that state.
Several repair attempts were necessary.
The aim was certainly not to wipe everything thoughtlessly.
We needed to try to preserve:
- personal files;
- software;
- settings;
- the working environment.
A conventional reset that retained the necessary elements did not work.
The system simply refused to return to a normal state.
Hours of Prompts, Research and Diagnostics
That is where artificial intelligence genuinely helped me.
It took hours of research, prompting, diagnostics and preparation.
Eventually, we arrived at a strategy that could reconstruct what was required to repair the system.
An additional problem was that the Windows version concerned was no longer readily available in its original form.
We therefore had to reconstruct a repair solution suited to the situation.
I am deliberately omitting the most technical details.
What matters here is not the exact sequence of commands.
It is the outcome.
Repairing Windows without Removing the Working Environment
I eventually managed to reinstall and repair Windows while retaining the necessary files, settings and applications.
From that point onwards, everything became much simpler.
Several versions of the system still needed to be upgraded.
Some upgrades need to be performed progressively rather than attempting to skip several stages at once.
Once you start again from a sufficiently healthy system, maintenance becomes much more conventional.
That makes all the difference.
A computer that is maintained regularly is far easier to manage than a system on which several years of problems have accumulated.
Two or Three Years of Maintenance Caught Up in One Week
In the end, I spent about a week on this machine.
That week made it possible to catch up on what I estimate to be two to three years of overdue computer maintenance.
For a first experience of this scale on such a degraded workstation, I find the result interesting.
And I can say it clearly:
without artificial intelligence, I do not think I would have managed to solve this problem in this way.
That does not mean AI did everything by itself.
It was necessary to:
- analyse;
- test;
- understand failures;
- change strategy;
- verify results;
- start again;
- retain what worked.
Above all, AI made it possible to retrieve a huge amount of technical knowledge and apply it progressively to the real problem.
Now I Am Experimenting with Codex on a Phone
Once that work was complete, I became interested in another subject.
What can be done directly on a phone with this type of tool?
I therefore began experimenting with Codex on my work phone.
This is obviously not the same as simply opening ChatGPT in a browser.
You need to be able to install and use technical tools that are not necessarily offered directly through the Play Store or App Store.
That still requires some knowledge.
But my objective is precisely to understand what is becoming possible.
I want to study my phone to determine:
- what I can analyse;
- what I can improve;
- what I can automate;
- what I can test;
- and which services might eventually be delivered remotely.
Phone, Computer, Tablet: We All Face the Same Organisational Problem
This reflection ultimately extends far beyond maintenance alone.
Today, almost everyone has files scattered everywhere.
There is:
- what is stored on the phone;
- what is stored on the computer;
- backups;
- photos;
- documents;
- different email addresses;
- online accounts;
- files that can no longer be found.
And, to be clear, it quickly becomes something of a mess.
How do we recover the data?
How do we connect the different devices?
How do we classify documents properly?
How do we know what already exists?
How do we retrieve information several years later?
All of this ultimately represents a genuine computer organisation task.
It is an area in which artificial intelligence could become particularly useful.
Using Codex Also Involves Risks
There is nevertheless an important limitation.
Tools capable of executing commands directly can also make mistakes.
If the request is poorly formulated, if the initial reasoning is wrong or if an operation is applied without sufficiently checking its consequences, problems may arise.
Not everything is authorised or feasible, of course, but that does not mean mistakes are impossible.
We therefore need to retain:
- rules;
- backups;
- checks;
- limits;
- human approval when necessary.
The more powerful a tool is, the more important it becomes to understand the framework in which it is being used.
New Generations Will Probably Go Much Further
I return finally to my opening question.
I do not know what young people who have already been taught about artificial intelligence think of all these experiments.
I am convinced that some of them are already going much further.
Some can control tools from their phones simply by speaking.
Others can produce mobile applications extremely quickly.
These are still areas that I am continuing to explore personally.
For my part, with the tools and experience currently available to me, I can already create certain websites and web applications within twenty-four to forty-eight hours when a project has been prepared sufficiently well.
What would have required much more time only a few years ago is therefore becoming accessible far more quickly.
Two Approaches to Artificial Intelligence
To summarise, I once again see two ways of approaching the subject.
1. The Technical Approach
You learn to understand the tools.
You work with:
- systems;
- code;
- commands;
- servers;
- files;
- automations;
- connections between services.
This approach requires more knowledge and learning.
In return, it generally enables you to understand more of what you are doing and retain greater control.
2. The Functional Approach
The second possibility is to use ready-made applications.
They can sometimes perform a large proportion of the work without requiring as much technical knowledge.
But this simplicity often has a cost:
- more AI credits;
- more subscriptions;
- certain limitations;
- dependence on the platform;
- sometimes a website or application that you do not control completely;
- recurring maintenance or operating costs.
Neither approach is automatically better than the other.
Everything depends on what you want to do, the time available, your knowledge and your budget.
What Ultimately Matters Is Not Just the Tool
Artificial intelligence is developing extremely quickly.
It already makes it possible to do things that would have seemed very complicated to me only a few years ago.
But my current experience leads me to a fairly simple conclusion:
artificial intelligence does not erase computing. It allows us to use in a different way the computing knowledge accumulated over decades.
There are extraordinary recent tools.
There are also old methods that remain extremely effective.
We often need to use both.
AI can help repair a computer, organise thousands of files, create an application, produce a website or analyse a phone.
But we must always begin with a real need, study the problem and prepare what we want to achieve.
That is probably where the genuine change lies today.
We are no longer merely using computers.
We are beginning to be able to talk to them and ask them to study, organise and carry out part of the computing work with us.