How artificial intelligence, Codex, AI agents, a master prompt, Git and a web server can help turn a simple idea into a first testable web application.
So, how can you develop a web application almost automatically with artificial intelligence?
Today, when you have access to artificial intelligence and to a web server with a remote terminal, you can already go very far, and much faster than before, in this kind of setup.
Of course, there are prerequisites. If you do not have access to a terminal, some possibilities will not work in the same way. There may also be restrictions depending on your hosting provider, your plan and the features available to you.
But broadly speaking, when you have a server rather than only shared hosting, or at least a sufficiently open environment, you can now connect artificial intelligence tools to the development of your project and begin automating many technical tasks that used to require specialists across several areas of computing.
Personally, I use OpenAI Codex.
For other services such as Gemini or Claude, I do not know their server-side possibilities well enough to discuss them precisely, even though I imagine that similar approaches exist through conversational agents designed for code and computer commands.
So here, I will mainly talk about what I have tested myself.
Start with the Objective, Not Necessarily with Technical Vocabulary
The first interesting thing about artificial intelligence is that you do not necessarily need to know all the technical terms immediately.
When you have at least a few basic computer skills, you can start by describing your project simply.
What do you want to produce?
What is your objective?
What should the application be used for?
What results do you want to obtain?
From there, ChatGPT or another conversational agent can already help you develop the broad ideas behind the project and identify the technical terms you need.
And if you do not know those terms, you can simply ask:
What technical terms should I use to build something simple and effective that matches this objective?
Artificial intelligence can then help you translate an idea expressed in everyday language into a structure that is easier to understand in a software development environment.
That is an important shift.
Instead of spending weeks learning all the vocabulary before you can even explain your project, you can begin by explaining what you want to achieve.
Then artificial intelligence helps you clarify the technical part more precisely.
In a way, you become the project manager of your own web application.
VPS or Shared Hosting?
Still, it is important to distinguish between a private server and shared hosting.
With shared hosting, some features may be limited by the hosting provider.
You therefore need to study what is actually available: terminal access, installable software, granted permissions, processes that can run on the server, and so on.
With a private server, you usually have much more technical freedom.
But that freedom has a counterpart: you also have to administer and secure your server.
And that does require more computer knowledge.
Your conversational agent can help demystify this part quite easily depending on your private server provider.
Personally, if someone knows almost nothing about system administration, I would not necessarily advise them to start immediately with a private server simply because it sounds more powerful.
Shared hosting, even with some limitations, can be more than enough for many personal and professional uses.
With the computing power available today, in 2026, you can already do a lot with relatively modest environments.
An entry-level private server may also cost around ten euros per month depending on the provider, but again: you need to know what you want to do with it.
From a Few Sentences to an Agent Structure
In an agentic conversation, I started from a few sentences describing an objective.
I asked ChatGPT to develop that idea and propose a structure.
From that structure, I then asked Codex to implement different agents.
That is where the logic becomes interesting.
You start with an idea expressed simply.
The conversational agent helps you transform that idea into several more precise components.
Then Codex can develop those components as agents that remain coherent with your project.
Of course, you can ask for very complex agents if you have the AI credits, resources and plan required.
But you can also begin with much lighter and more concise agents.
In my opinion, that is even a good way to start.
Why try to build something huge immediately if a small first version can already show whether the idea works?
The Role of the Master Prompt
Once the agent structure was in place, I asked for the creation of a master prompt.
Here, the master prompt is used to coordinate the work according to the precise structure of the agents.
After that, the logic becomes relatively simple.
You can ask Codex to save this master prompt inside the project and use it as the framework for implementing a first version of the application.
For example:
you define the agents;
you define their roles;
you build a master prompt consistent with that architecture;
then you ask for a lightweight first implementation matching what has been defined, in order to obtain a functional mock-up.
In my demonstration, I quickly reached a lightweight first production version, in record time compared with my initial level of knowledge.
Obviously, I am not talking here about a huge platform containing years of development.
It was a modest application.
But that is precisely what makes it interesting.
A First Application in a Few Hours
In my case, the initial structure and the agents required about one hour of work.
Then Codex made it possible to continue the implementation, coding and setup.
When I looked at the result, the work had been completed much faster than one might have expected only a few years ago.
Basically, in a few hours, I had a small first web application matching the strategy I had just described.
Once again, this does not mean that a complex, secure, tested application perfectly suited for production can be built automatically in two hours.
But if the goal is to obtain a functional mock-up or a first version that can be tested quickly, the possibilities become extremely interesting.
And I think this is where the shift needs to be understood.
We are no longer simply talking about a hypothetical future.
These tools can be used today.
Then You Need to Version the Project with Git
Once a first version exists, you need to be able to track its evolution.
That is where Git comes in.
Git is a version control tool used very widely in software development.
It makes it possible to track file changes, group them into versions, work with different branches and keep a clean history of the development process and its current state.
Git has a reputation for seeming complex when you first start using it.
I was joking about this in the video, in relation to the name “Git” and the humour of its creator, Linus Torvalds. In English, “git” can mean an unpleasant or foolish person.
But humour aside, version control is extremely important.
What changes with artificial intelligence is that you can now ask an agent to help you manage this complexity.
For example, my next prompt can simply ask it to create appropriate versioned groups of files in order to obtain a clean Git tree.
To put it simply, a commit corresponds to a coherent state in the evolution of your project: a group of files at a given moment in development, testing, pre-production or production.
This prevents you from modifying dozens of files manually without being able to go back, or without knowing when it is time to test, refine or validate the work.
Working Through Successive Passes
This is also a way of working with artificial intelligence that I find interesting.
I can ask for a first exploratory pass for a mock-up.
Then I test.
Does it work?
Does it match my expectations?
Is the initial objective respected?
If not, I can ask for a second pass using another approach, or ask for a deeper analysis.
And between these steps, I can version the project.
So the process becomes something like this:
a first idea;
a first structure;
a first implementation;
a test;
version control;
then a new pass.
You move forward progressively.
This is not necessarily very different from the historical principle of software development.
What changes enormously is the speed at which some tasks can now be analysed, coded, corrected or documented.
Artificial Intelligence Does Not Remove the Need for Computer Skills
Still, we need to be careful with one shortcut.
Saying that artificial intelligence can code does not mean that computer skills are no longer needed, even though ChatGPT can simplify a very large part of the work.
For example, properly administering and securing a private server takes time and requires at least a minimum understanding of what you are doing, or of what you need compared with the tools available on shared hosting or an online hosting service.
The further you want to go, the more you need to understand the actions to be carried out as the project manager of your web project.
That is also why I would recommend starting with an environment adapted to your current skills.
If you have shared hosting that gives you access to a sufficiently open remote terminal, that may already be interesting.
If you have more experience, you can then go further with a private server.
And if you do not know how to connect your server to your artificial intelligence environment, you can already ask your conversational agent to explain the different steps adapted to your situation.
Let Us Focus More on What We Want to Produce
What interests me most in this evolution is the shift in the problem itself.
For a long time, to create something in computing, you first had to master a large amount of technical vocabulary and complete programming languages.
Today, that vocabulary can be explained or translated by artificial intelligence.
This allows us to focus more on another question:
What do I really want to produce?
What results do I want to obtain?
Who will this application serve?
What problem do I want to solve?
Technical vocabulary remains useful.
And today, we can ask artificial intelligence to help us understand it at the moment when we need it.
A Phone Is Already a Small Computer
It is also useful to put things into perspective.
Today, a phone is already an extremely powerful small computer.
Depending on the tools available, you can think through your project from a computer, a tablet or even a phone.
Of course, not every device is equally practical.
But access to computing tools and artificial intelligence has become much more democratic.
And that is what makes all these possibilities interesting.
Creating an application no longer necessarily begins with installing an extremely complex environment on a huge machine.
You can start with an idea, a basic device, an internet connection and AI credits in an affordable plan.
Being a Computer Professional Does Not Only Mean Coding a Web Application
I also think we need to broaden the idea of computing.
Being a computer professional does not only mean coding web applications or software.
Anything related to the automated processing of information can now be connected to many other uses.
You can create agents to work on a book.
You can create agents to produce or organize presentation documents.
You can imagine transforming the content of a video channel into a structured base that can later be used to work on a book.
You can create proofreading agents.
Correction agents.
Agents specialised in reference research.
Tools for working on indexes, glossaries or the structure of long documents.
And then there are audio, video, images and text generation.
The possibilities are extremely varied.
Specialised Agents Rather Than an AI That Has to Guess Everything
This is also why I find the idea of agents interesting.
Instead of asking one single prompt to do absolutely everything, we can imagine several specialised agents.
One agent has a precise role.
Another agent has another function.
Then a master prompt organizes the work.
The conversational agent can help you define this structure.
Then the development-focused artificial intelligence can progressively implement the components.
Finally, the master prompt can coordinate the whole process until a first usable version is obtained.
Once again, the more complex the project becomes, the more you need to test and understand what you are building.
But for creating mock-ups, prototypes or quickly exploring several ideas, I find this approach particularly interesting.
AI Also Changes Our Relationship with Technical Knowledge
This evolution inevitably raises a question about learning.
Do we still need to learn every technical detail when artificial intelligence can help us use them?
In my opinion, we probably need to distinguish between two things.
There is what we want to accomplish.
And then there is the level of understanding required to check that we are not doing anything careless.
We can delegate part of the vocabulary and part of the technical work to artificial intelligence.
But we still need to understand the field well enough to define the right objectives, recognise a problem and verify the results.
This is especially true when security, data or a server accessible from the Internet are involved.
AI can accelerate the work enormously.
It does not automatically turn every decision into a good decision.
We Are Already in This New Phase of Web Development
In the end, what I find most interesting is the speed with which we can move from an idea to something testable.
An idea can be described in a few sentences.
Artificial intelligence can help find the right technical terms.
That idea can be transformed into an agent structure.
A master prompt can be created.
Codex can help implement a first version.
Git can version the project.
Then we test, correct and start again.
Only a few years ago, many of these steps required much more time or several different skill sets.
Today, one person equipped with the right tools can already experiment with a great many things.
This does not mean that developers, system administrators or other IT professions no longer matter.
On the contrary.
The more powerful the tools become, the more important it is to know what we really want to do with them.
So, if I had to summarise my approach, I would simply say this:
Let artificial intelligence help you with technical complexity, and focus above all on the result you want to produce.
Start small.
Build a first version.
Test it.
Version it.
Then improve it progressively.
I will stop here for this demonstration.
José DUTIAN