A new study has found that nearly a third of code is now generated by artificial intelligence, and that there is an AI divide in programming: experienced developers are becoming more productive, while junior developers are not.
A new study has found that nearly a third of code is now generated by artificial intelligence, and that there is an AI divide in programming: experienced developers are becoming more productive, while junior developers are not.
Over the past two years, the global volume of AI code written has increased sixfold. While in 2022 this figure was only 5%, by the end of 2024 it reached almost 30%, writes ZDNET.
The findings are reported in a study by the Complexity Science Hub (CSH), which examined the impact of AI tools and platforms on software development. The team developed a model and applied it to a large dataset covering software development in six countries.
According to the study, AI has increased programmer productivity by almost 4%. However, there is a gap in the data: less experienced programmers use AI more often (37%), but the productivity gains are observed almost exclusively among experienced developers.
“By comparing the same developer before and after the implementation of AI generation, we show that the implementation of AI significantly increases productivity. Developers using AI generation are also more likely to include new combinations of software libraries in their code, indicating that they are venturing into new technical areas using unfamiliar building blocks,” said Simone Daniotti, lead author of the study.
However, the CSH study found a striking lack of productivity among early-stage developers. Rates of generative AI adoption are higher among early-career developers. However, the gains in both productivity and depth of research are concentrated almost exclusively among senior-level developers. In contrast, early-stage developers, while using generative AI more frequently, do not reap the same benefits.
The researchers believe this is due to developers' ability to apply AI to complex tasks. Senior developers, for example, are faster at interpreting and detecting errors in code generated by AI.
The CSH study also showed that thanks to AI, developers have become more likely to experiment with new libraries.
“This suggests that generative AI has enabled users to advance more quickly into new areas of programming, introducing new types of functionality into their code,” the researchers say. “Generative AI drives individual innovation, expanding the capabilities of specialists to use new combinations of libraries. However, again, only experienced senior-level developers are able to use AI in a way that has important implications for career development and learning in the age of artificial intelligence.”
Another study found that American programmers most often use AI to write code .
