US-based software developers are the world’s most prolific users of AI coding assistants, a trend that researchers believe has national economic implications.
A quartet of researchers explore American coders’ liking for helpful bots in a preprint paper that analyzed 80 million code submissions to GitHub between 2018 and 2024.
The researchers – Simone Daniotti, Johannes Wachs, Xiangnan Feng, and Frank Neffke – devised a machine learning model to analyze GitHub submissions and found an estimated 30.1 percent of US-sourced Python functions submitted to GitHub in 2024 were AI-generated.
Germany came in next at 24.3 percent, ahead of France (23.2 percent), India (21.6 percent), Russia (15.4 percent) and China (11.7 percent).
The paper argues that once developers use AI for 30 percent of their code, quarterly commits rise by 2.4 percent.
“Coupling this effect with occupational task and wage data puts the annual value of AI-assisted coding in the United States at $9.6–$14.4 billion,” the authors argue.
The estimate aligns with Microsoft CEO Satya Nadella’s claim that presently about 30 percent of Microsoft code is written by AI.
The potential economic benefit arising from AI-enhanced commit rates could be even higher, the authors argue, if higher productivity claims from other AI surveys are used, such as this report from last September that found a 26 percent productivity improvement.
Based on task completion time estimates from three different randomized controlled trials that found productivity improvements of 16.5 percent, 6.3 percent, and 26 percent respectively, the researchers conclude 30 percent AI usage would lead to a productivity lift worth between $64 billion and $96 billion annually.
The authors acknowledge that their estimates have limitations. They note, for example, that their focus on GitHub code submissions may miss those made to Gitee, which is popular in China. And they say they don’t take into account “any potential reduction in value of coding tasks due to additional supply of code by AI.”
There are other factors that may skew the author’s results, like treating Python as representative of the impact on software development in other languages and assuming that the rate of AI usage in open source projects on GitHub is repeated in other settings.
But overall, the authors say they’re bullish on the productivity value of AI. What’s more, they say, AI adoption leads to increased experimentation with new software libraries and combinations of libraries, which expands developer knowledge. That assumes those libraries actually exist and weren’t dreamt up by AI.
Outside of writing code, the economic impact of AI may be more modest. In a paper published last year, “The Simple Macroeconomics of AI,” MIT Institute professor Daron Acemoglu predicted AI-driven productivity gains of only about 0.7 percent. ®