AI writes more and more computer code

In the US, artificial intelligence writes as much as one third of programming code. However, it only brings benefits to experienced programmers, note the authors of the study in Science.

A sharp increase in the share of AI in writing code

As the authors of the study described in “Science” remind us, the programming industry is gigantic – in the United States alone, companies spend $600 billion on remuneration of specialists in this field.

As a new analysis has shown, by the end of 2024 in the USA, every third newly written function, i.e. an independent subprogram in the code structure, was created with the support of artificial intelligence systems.

The share of artificial intelligence increased from only 5%. in 2022 – to nearly 30 percent in the last quarter of 2024.

“We analyzed over 30 million contributions written in Python from nearly 160,000 developers using GitHub, the world's largest platform for collaborative code creation,” explains Simone Daniotti from Complexity Science Hub and Utrecht University.

As scientists explain, GitHub records every stage of work – from adding a new line of code, through editing, to optimization. Thanks to this, you can track the work of programmers from all over the world in real time. Python itself remains one of the most popular programming languages ​​in the world.

Big differences between countries

At the same time, the study revealed significant differences between individual countries. While the share of code created with AI support is highest in the US (29%), Germany reaches 23% and France 24%. Right behind them, with a score of 20 percent, is India, which is rapidly catching up.

In turn, Russia (15%) and China (12%) were still clearly lagging behind at the end of the research period.

“It is no surprise that the USA is the leader – this is where the leading language models (LLM) come from. Developers in China and Russia have had to face barriers in access to these technologies, blocked either by their own governments or by suppliers, although VPN workarounds are still used. Recent Chinese developments, such as the DeepSeek model released after the end of data collection, in early 2025, suggest that this gap can be quickly closed,” notes Prof. Johannes Wachs, one of the authors of the study.

AI has increased developer productivity

The study shows that the use of generative artificial intelligence increased developer productivity by 3.6% by the end of 2024. This means a profit of USD 23 to USD 38 billion per year, researchers calculate.

“Although this may seem a modest result, on the scale of the entire global software industry it means a huge profit,” emphasizes one of the experts, Prof. Frank Neffke.

Test showed no gender differences in AI use.

However, seniority turned out to be crucial. Well, less experienced programmers rely on artificial intelligence to create as much as 37%. their code, while for industry veterans this indicator is only 27%.

However, experience is fundamental to the benefits obtained. “Beginners have almost no benefit from this. Generative artificial intelligence does not automatically equalize opportunities; on the contrary, it may even deepen existing differences,” says Daniotti.

Additionally, experienced developers are more likely to experiment with new libraries and unusual combinations of available tools.

“This is proof that artificial intelligence not only automates routine tasks, but also radically accelerates the learning process. Thanks to it, experts can expand their competences more efficiently and boldly enter completely new areas of programming,” notes Prof. Wachs.

Scientists predict an increasingly stronger impact of AI on programming. “For business, decision-makers and educational institutions, the key question is no longer 'whether to use AI' – but how to make its benefits available to everyone, without deepening inequality,” says Prof. Wachs.

“In a world where even a car has become an IT product, we must identify barriers to the implementation of artificial intelligence as quickly as possible: from the level of individual companies to entire regions and countries,” adds Prof. Neffke.

Marek Matacz

mate/ zan/