AI in software dev: Billions in value added, study shows
AI-assisted programming is becoming increasingly prevalent. However, only experienced developers achieve productivity gains.
(Image: Shutterstock/Alexander Supertramp)
- Manuel Masiero
In the US software industry, almost 30 percent of Python code is now written by AI. This is the conclusion of an analysis by the Complexity Science Hub (CSH) in Vienna in collaboration with Utrecht University. It covers the period from early 2019 to the end of 2024. Europe follows in second place for usage.
The study, published in the journal Science, shows that there are significant regional differences in the acceptance of AI-written Python code. It is highest in the USA with a share of 29 percent. In 2022, the figure there was still around five percent. France achieves a share of 24 percent, followed by Germany with 23 percent and India with 20 percent. Russia (15%) and China (12%), on the other hand, lagged significantly behind at the end of the study.
The data basis for the study was also determined with the help of AI. A specially trained AI model was used, which analyzed more than 30 million Python code blocks from around 160,000 developers on GitHub. The AI model checked whether the Python contributions were created, for example, via ChatGPT or GitHub Copilot.
(Image: Complexity Science Hub Vienna (CSH))
The less experienced, the more AI
No differences were found between women and men in the study regarding AI usage. However, experience plays a major role. While less experienced programmers used generative AI in 37 percent of their code, the share among experienced clientele was only 27 percent. This hardly helped the former, as productivity increases were only observed in the latter. However, precise classification is difficult because developers use AI tools in very different ways.
Videos by heise
According to the CSH, the use of generative AI increased programmer productivity by 3.6 percent by the end of 2024. While this sounds like little, it represents significant value creation. For example, an estimated 637 billion to 1.06 trillion US dollars are invested annually in programming tasks in the USA. A productivity increase of 3.6 percent therefore corresponds to an additional value of between 23 and 38 billion US dollars per year. However, the study qualifies that this value generalizes the measurements of Python code.
(mho)
