AI writes the code, but the project remains a matter of experience

Those who have worked in development for a few years know well: the hardest part of a project is rarely writing lines of code. The real work happens before, in understanding what the client really needs, how to organize data, which parts of the system will need to grow and which will remain stable. And it also happens after, in maintenance, updates, and managing errors that only emerge when the project is in production and people start using it.

AI tools have made the central phase, writing, extremely fast. Today, just describing a function is enough to get it in seconds. It's a huge change and it would be foolish to underestimate it, so much so that every developer now makes extensive use of it. However, it has also created a widespread misunderstanding, the idea that anyone, with the right prompt, can build a solid application.

Working well with AI is much like coordinating a very fast and tireless collaborator who never says "this can't be done." However, it doesn't know the project and doesn't bear responsibility for it. To achieve results, you need to know how to do some things that you can't learn in a few months or, worse, only through vibe coding experiences.

Break down the problem. An experienced developer divides an application into clear parts and assigns AI limited tasks with precise constraints. They know which decisions to keep for themselves and which to delegate.

Read the generated code with a critical eye. AI proposes, but someone must evaluate. Recognizing a fragile solution, an inefficient query, or a pattern that will cause problems in the future requires having seen and experienced them often firsthand (like the hours, days, and weeks spent on Stack Overflow hoping to find solutions to problems already faced and solved by other developer colleagues).

Know the architecture. Choosing how to structure data, where to put logic, how to make frontend and backend communicate, how to manage cache and performance. These are decisions that determine a project's fate and that AI executes well only when someone has made them with full knowledge.

Know when to stop. Sometimes the best answer is not to write new code but to use something that already exists or simplify the client's request. This kind of judgment comes from experience, not from a prompt.

In practice, AI multiplies the abilities of those who use it. If the starting skills are solid, the result is impressive. If they are weak, errors multiply as well.

An AI model responds to what it is asked. If the request is vague or poorly set, the result will be consistent with that request, thus convincingly wrong. The code works, the page loads, the form sends data. Everything seems fine.

Problems arise weeks later. A booking module that doesn't handle two simultaneous requests. A database built without considering how contents will be filtered when there are thousands. Credentials left in the code. Dependencies added without reason that slow down the site and open security holes. Those without experience don't notice anything because they don't know where to look.

There is also another, more subtle effect. Every AI response is built on the context available at that moment. Without someone maintaining an overview, the project becomes a set of pieces that work individually but don't communicate well with each other. Adding a new feature becomes increasingly costly, to the point where it is better to start over from scratch.

The American job market, which often anticipates European trends, is showing this phenomenon very clearly. Tech companies have reduced hiring junior profiles and focus their search on senior developers capable of designing systems and guiding the work of AI tools.

The logic is simple. Many tasks once assigned to juniors, such as writing repetitive code, basic testing, or small fixes, are now largely performed by AI. What cannot be automated is the ability to decide what to build and how. A senior with the right tools can cover a workload that previously required a small team, maintaining control over quality.

For companies, this means faster projects without sacrificing solidity. For the sector, it means an open question: if juniors struggle to enter, who will become senior in ten years?

What happens in code has a clear parallel in the design world. Since generating interfaces, images, and layouts with AI has become accessible to everyone, the web has filled with what the industry calls slop: mass-produced content, visually clean but empty, all the same.

You can recognize it at a glance. Landing pages with the same structure, the same gradients, the same rounded cards, and the same icons. Generic illustrations that could be on any site. Neat dashboards but designed without asking who will use them and for what purpose. The result is superficially correct but communicates nothing about the brand and often does not help the user achieve their goal.

The reason is the same as we saw for development. AI tends to propose the most probable solution, that is, the average of everything it has already seen. An experienced designer uses this base as a starting point and then makes decisions: a visual hierarchy built on real content, typography chosen for the brand's tone, navigation flows designed around user behaviors, details that make a product recognizable. Those who improvise stop at the first proposal because they lack the tools to understand what is missing.

In digital projects, design and development are increasingly intertwined. An interface generated without criteria brings usability, accessibility, and consistency problems that then affect code and maintenance costs. In both cases, AI accelerates the work of those who already know what they are doing and produces mediocre results in the hands of those who do not.

This transformation also closely concerns those who commission a digital project. The market is filling with offers at very low prices, made possible by the speed with which code is produced today. A website or an application can be delivered in a few days and appear perfect at the first demo.

The difference shows over time. A project built without real design becomes difficult to update, integrate with other services, and grow. The costs saved at the start return in the form of urgent interventions, reworks, and security problems.

When evaluating a supplier, it is advisable to ask questions that go beyond timing and price. How will the project be structured? Who is responsible for code review? How are updates and security managed in the long term?

Artificial intelligence is the most powerful tool to have arrived in software development in many years. In our daily work, we use it every day and it allows us to focus time on decisions that really matter. Precisely for this reason, we know that its value depends on who guides it.

It is not enough to have access to the tool. You need to know what to ask, how to verify what it produces, and how to integrate it into a project designed to last. It is a skill built with years of projects, mistakes (frustrations), and solutions. And that is why today, paradoxically, experience is worth much more than before.