Artificial intelligence and web development: between real progress and illusions
Over the past two years, artificial intelligence has taken the digital sector by storm at a remarkable pace. In software development, code assistants, automatic generators and conversational models are now part of the daily routine for many developers.
The question, then, is no longer whether to use AI. The real question is rather: how to use it wisely.
At WebstanZ, we have gradually incorporated these tools into our working practices. But we do so with a firm conviction: AI is a powerful tool for enhancing human work, not a substitute for software engineering. And certainly not in critical projects.
AI in teams’ day-to-day work
In its simplest form, AI currently acts as a development assistant. A developer can submit an obscure error that appears in the logs, ask for suggestions on how to refactor a function that is a little too complex, or quickly generate a unit test to verify a specific behaviour.
In these situations, AI plays a role quite similar to the one Stack Overflow used to fulfil: a quick way to explore possibilities and save time when solving a problem.
But the responsibility clearly lies with the developer. It is the developer who understands, validates and adapts the output.
Taking it a step further, AI can also act as a genuine development co-pilot. It can suggest implementations, propose optimisations or help generate an initial code structure. In some cases, this really speeds up the work, particularly when it comes to producing repetitive code, writing documentation or generating tests.
Provided, however, that the project is built on a solid architectural foundation and that the teams maintain strict discipline: code reviews, a thorough understanding of the components produced, and adherence to the project’s technical standards.
Because AI also has a formidable ability: to generate code very quickly… even when it isn’t suitable.
When AI becomes the lead developer
For some time now, a concept has been doing the rounds in the tech community : vibe coding. The principle is simple: instead of writing code, you describe the application in natural language and the AI generates the bulk of the solution. In certain contexts, this approach can be impressive. In just a few hours, it is possible to produce a working prototype that would have taken several days using traditional development methods.
But this approach quickly reaches its limits when it comes to real-world projects. A recent example brought this home to us: a company contacted us a few months ago after having an internal web app developed by an external contractor. The project had been delivered quickly, and the initial demonstration was convincing. The interface worked, the main screens were in place and the initial features seemed to meet their needs. But very soon, problems began to arise.
Certain features became unstable as soon as realistic volumes of data were introduced. Unexpected behaviour would occur under certain conditions. And, above all, when it came to scaling the system, no one seemed to really understand how it worked.
Upon analysing the code, the diagnosis became clear. The bulk of the application had been generated from prompts fed to an AI model. The project was not based on an architecture designed from the outset, but on an accumulation of code blocks generated as requests came in. The result resembled an extended prototype more than software designed to last. The system worked… as long as you didn’t push it too far.
But for an ERP application at the heart of the company’s operations, this level of fragility quickly became a major risk. In this specific case, vibe coding had perfectly fulfilled its role in rapid prototyping. It had made it possible to bring an idea to life and demonstrate an interface. But it had not produced a system capable of withstanding the test of time, workload and business changes.
The myth of the “10x” developer
In discussions about AI, we often hear the idea that developers could become ten times more productive. The reality is more nuanced. A recent post d’Iztok Smolic, from the agency Agiledrop, explains this very well by referring to a well-known principle in computer science: Amdahl’s law.
When estimating a six-month software project, his team realised that the time spent writing code was, in fact, only part of the work. A large part of the schedule actually depended on other factors: clarifying requirements, obtaining client feedback, coordinating decisions among stakeholders, conducting code reviews, fixing bugs and preparing deployments.
In other words, even if AI made writing code ten times faster, the overall benefit would remain limited, simply because code is just one element among many in the development of a digital project.
Indeed, some analyses show that the widespread adoption of AI in certain teams has had a paradoxical effect: more code generated, more pull requests… and therefore more review work. Productivity does not depend solely on the speed at which code is written. It depends above all on a team’s ability to make decisions, collaborate and maintain the quality of the system over time.
AI is a powerful tool… provided we maintain high standards
This obviously does not mean that AI has no place in software development. Quite the contrary.
When used wisely, it becomes a powerful tool for speeding up problem-solving, exploring solutions or reducing repetitive tasks. It allows developers to devote more time to what really matters: architecture, code quality, security and understanding the business.
But this effectiveness hinges on one simple condition: AI must remain a tool in the service of sound engineering. Because building a sustainable application is not simply a matter of writing code. You need to understand the requirements, structure the architecture, anticipate future developments, ensure security, and make sure that other teams will be able to maintain the system for years to come.
And above all, it remains a deeply human endeavour.
AI should serve projects, not the other way round
At WebstanZ, our approach is simple: to use artificial intelligence where it delivers real value, whilst maintaining the engineering standards required for project success. This means streamlining certain tasks where appropriate, but also upholding high standards in terms of architecture, security and maintainability.
Because, ultimately, a good digital project isn’t just one that works today. It’s one that will continue to work, evolve and be understood in five or ten years’ time. And no artificial intelligence can yet guarantee that on its own.
KEY POINTS FROM THIS ARTICLE
- AI is establishing itself as an indispensable tool in web development: AI code assistants and templates are now part of developers’ daily routine, but their use must be carefully considered.
- An assistant role that speeds up certain tasks : AI helps save time on debugging, refactoring, and generating tests or documentation, whilst remaining under human control.
- A real benefit… but one that must be guided by best practice : Effectiveness depends on a robust architecture, rigorous code reviews and a thorough understanding of the generated code.
- The limitations of “vibe coding” in complex projects : Generating an application using prompts may work for a prototype, but often results in fragile systems that are difficult to maintain and scale.
- The myth of the “10x” developer is overblown : Development is about more than just writing code: coordination, validation, quality control and decision-making limit the actual gains in productivity.
- AI must remain a tool, not a substitute for engineering : Sustainable projects rely above all on human decisions: architecture, security, maintainability and business understanding.