Study Reveals AI‑Assisted Coding Powers a Third of Software

Recent research published in Science shows that generative‑AI tools now generate roughly one‑third of all newly written software functions in the United States. By analyzing over 30 million Python contributions from more than 160 000 developers on GitHub, the study found AI‑assisted code rising from 5 % in 2022 to 29 % by the end of 2024.

Methodology Behind the AI‑Assisted Code Estimate

Data Set and Classification Approach

The Complexity Science Hub (CSH) and Utrecht University built a custom AI classifier to identify code blocks created or heavily assisted by large‑language‑model services such as GitHub Copilot, OpenAI Codex, Google Gemini, and Claude Code. The classifier was applied to the full commit history on GitHub, focusing on self‑contained Python functions to calculate the share of “new code” that involved AI assistance.

Impact on Developers by Experience Level

Productivity Gains for Senior Engineers

For experienced programmers, AI assistance correlated with higher productivity: increased commit frequency, broader use of software libraries, and more rapid exploration of new functionality. Senior developers can leverage AI suggestions to accelerate complex tasks and reduce routine boilerplate.

Limited Benefits for Junior Developers

Junior developers showed no statistically significant productivity improvement. The study suggests that less‑experienced engineers may lack the expertise to effectively integrate AI‑generated suggestions into their workflow.

Emerging “Vibe Coding” Workflow

“Vibe coding” describes an AI‑assisted workflow where developers issue natural‑language prompts and receive complete code snippets or full applications. By treating spoken language as the primary programming interface, developers can request, for example, “a React component that displays a sortable table of sales data” and obtain functional code within minutes.

Economic Implications for the U.S. Tech Industry

The United States spends an estimated $600 billion annually on wages for coding‑related work. If AI continues to generate a third of new code, the productivity boost for senior developers could translate into measurable cost savings and faster time‑to‑market for software products.

Geographic Adoption Patterns

Regional analysis shows the United States leading with 29 % AI‑assisted code, followed by Germany (23 %) and France (24 %). India is rapidly catching up at 20 %, while China (12 %) and Russia (15 %) lag due to limited access to leading LLMs.

Future Outlook Across Programming Languages

Although the current data focus on Python, the authors expect similar adoption trends in JavaScript, Java, C#, and other ecosystems. As LLMs improve in code understanding and debugging, the proportion of AI‑generated code is likely to increase further.