Artificial intelligence (AI) tools are becoming an indispensable part of the daily work of programmers.
From speeding up the process of writing code, learning new programming languages to handling simple tasks, platforms like GitHub Copilot or Cursor are present in most software development environments. However, their trust in them tends to decrease.
According to the latest annual survey of Stack Overflow (an online Q&A platform dedicated to the community of programmers and software engineers) with more than 49,000 programmers participating, up to 80% said they are using AI tools in their workflow.
However, only 29% believe in the accuracy of AI-generated code, down sharply from 40% last year. This difference shows a complex relationship between convenience and reliability.
The main reason comes from the nature of AI-generated code when they seem correct but often contain small, difficult-to-detect errors, especially dangerous for inexperienced programmers.
Also in the survey, 45% of respondents confirmed that near-correct answers were the biggest problem, making the troubleshooting process more time-consuming.
Many developers have returned to Stack Overflow to find answers after problems with AI code. More than a third of survey participants admitted to doing this regularly.
In addition, 72% of people opposed copying AI-insulated code without careful testing.
Despite the shortcomings, AI is still a useful tool if used properly. Many experts say that instead of considering AI as a perfect sub-pject, programmers should consider it as a trainer, helping to suggest ideas and support thinking and not completely replacing it. This approach is not only safer but also makes the learning process more effective.
Even Stack Overflow, a platform that was once worried about "dying" because AI was also having to adapt. Although traffic has decreased, the platform representative affirmed that this is not too serious. Instead, they are seriously re-evaluating the criteria for success in a rapidly changing digital era.
