Artificial intelligence has also crept into the sector of programmers, giving rise to new ways of working such as the so-called vibe coding.
This technique consists of using ChatGPT and other AI in a completely natural and spoken way to generate code fragments or complete programs with just a few instructions. The idea is to save time and avoid thinking so much about overly technical and tedious details.
Now, instead of being stuck writing lines and lines of code, developers can become managers, project designers, and supervisors of the quality of the code generated by machines.
However, all that glitters is not gold. As explained by a programmer in The RegisterFor him, programming is still a task that requires critical thinking and care so that the code is readable and can be maintained over time. AND Although he uses AI tools from time to time, he acknowledges that the results can be vague or even full of errors.
«AI tools always give answers with certainty and forcefulness. Even with poor or incomplete answers, they leave no room for doubt! Anyone with some experience knows that even the best plans sometimes fail, that life is never so certain. Maybe it’s just my impression, but the cheerful assurance of AI tools is irritating… and worrying,» he says.
The enormous myth that AI in programming saves you time
With this trend on the rise, there are many who claim that these assistants work almost miracles, saving the programmer a lot of time. While this is true in some minimal cases, the engineer makes it clear that there are many nuances.
Clarifies that, although for very simple or repetitive tasks AI can give you good results and speed up parts of the work, when it comes to code for more complicated applications or that must be maintained for years, this changes.
«I asked CoPilot how to read the application associated with PDFs from the registry «in Delphi». CoPilot confidently suggested that I just look in «HKCU\Software\Classes\.pdf». Although the code didn’t compile in my version of Delphi, the real problem was that it was missing details. Finally, I found a question on StackOverflow (see here) that explained the subtleties in more detail,» he says.
For him, the advice is simple: AI should be a tool to complement human work, not replace it. The programmer must be in the driver’s seat, understanding the code and making decisions.
AI should never be blindly trusted. The human hand is vital to avoid cybersecurity problems
This is something that Adam Conway, of XDA Developerswho wanted to test how good AI is at programming and has no doubt: it is horrible.
When reviewing the code generated by ChatGPT, especially using GPT-5, he found all kinds of bugs that can be a serious problem if implemented in an application. In his opinion, vibe coding demonstrated that without supervision, AI can produce code with vulnerabilities.
For example, in a program for reporting system statistics, authentication was weak, with passwords sent in plain text and no real protection. In addition, it explains that input controls were missing that could saturate or completely collapse servers with denial-of-service type attacks.
But this is not all because the programmer ran into even more serious problems. In some cases, the code allowed a user with access to modify data or inject malicious commands..
In another example, a directory listing program in Windows was vulnerable to command injections that could run applications such as the calculator. Furthermore, in a simple web server written in C generated by AI, flaws were detected that could allow access to parts of the system to also take control.