AI Code Detector is powered by span-detect-1, a machine learning classifier trained on millions of AI-generated and human-written code samples. It looks at semantically-generated "chunks" of code to classify them as AI-generated and human-written based on patterns in style, syntax, and structure.

AI coding tools are everywhere, but it's very difficult to understand their actual impact. Vendors sometimes have reporting systems, but they are inconsistent with one another. The machine learning model that powers AI Code Detector, span-detect-1, gives engineering leaders real visibility into the AI code in their repos.

AI Code Detector’s underlying model, span-detect-1, is 95% accurate for Python, TypeScript, JavaScript, and Ruby. Accuracy varies with chunk size, but we consistently test on independent datasets to avoid overfitting and to keep ourselves honest.

Some code, like import statements and other boilerplate, doesn't contain enough signal to classify confidently. Rather than guess, span-detect-1 abstains and marks those chunks as unknown. About 10% of chunks fall into this category.

Yes. You can use span-detect-1 as part of Span’s developer intelligence platform. With Span, you can not only detect AI- versus human-authored code, but also track adoption KPIs, quality outcomes, and the impact of AI coding assistants.