Using Artificial Intelligence to write code: a cliché topic?

I constantly hear managers, tech leads, influencers, and even coworkers repeating the same phrase to exhaustion:

"We no longer need to write code. AI writes code far better than we do."

For a few months, I truly believed that premise and kept an open mind as I took on side projects to test the hypothesis. Early on, I felt an absurd increase in speed. Tasks were delivered more quickly, the product team was pleased, and they believed that more adjustments and new features could reach the engineering team. I had the opportunity to build a project from scratch for a large audience, in which the entire development team relied heavily on tools such as Cursor, Claude Code, and Codex. As the project grew, some issues began to bother me.

However hard I tried to take notes, draw diagrams, and understand the business rules, I sometimes found myself lost in the code, trying to discover where the rule that governed a calculation or the configuration of a variable lived. I validated that feeling in conversations with the team and noticed a pattern: no one knew the code particularly well. When we needed to debug production issues, we were entirely dependent on AI to understand and locate sections of the code. It felt as though I was no longer the owner of the code and, consequently, of the product. Over time, I realized that the more the project grew, the more the team's insecurity grew regarding the implementation of new features or refactorings. The team feared the code, as well as interacting with it.

Another source of discomfort was that all the productivity we gained at the beginning was lost. Tasks, debugging sessions, and refinements started taking longer. Naturally, humans do not retain context the way AI does, and even though it had produced the code, it often proved inefficient at answering questions about code it had written itself. In the end, I felt that, although it is a powerful tool for helping write documentation and tests, I was not convinced to hand over all of my work to AI. The core of the application, the domain, and the business rules require a thoroughly human perspective. Perhaps I have not yet learned how to work at that stage with AI as a partner.

To make matters worse, I sense pressure in the industry for us to be literally pushed into "the new way of programming." It strikes me more as the desperate attempt of a market that cannot sustain itself by manufacturing demand for a technology that has proven efficient, but not for everything it claims to do, nor at the volume and scale at which it is sold in countless advertisements. For building POCs, it is truly an outstanding tool. Even so, I believe that delegating the creation of all code to AI is a long-term mistake.

On the other hand, some tasks that once took a great deal of a developer's time, such as picking up a legacy project and understanding it, can now, with a few prompts and careful reading, make it possible to understand a project thoroughly in a matter of hours. I consider this the greatest gain, especially for developers who are just starting in the profession or who have joined a new organization.

I hope to keep studying and improving my skills in AI-assisted development. Yet I feel that I, like some other developers, am going through a period of mourning, given that a very enjoyable and rewarding part of the profession is dying. The pleasure of sitting down and writing code that you thought through no longer exists.