This website checks git repositories and hex packages for signs of AI-led development. It doesn't try to analyse the code itself, but instead scans for metadata that might indicate heavy AI use.
Currently, it checks for commits co-authored by known AI agents, as well as common configuration files for tools like Copilot, Cursor, Claude, and others.
While there are some ethical and moral concerns, this project does not exist to shame anyone. Please be respectful.
What has changed, obserably, is code quality. GitHub struggles to be fully operational more than 85% of the time now. Of course, Microsofts push towards AI-driven development isn't the only reason for that and in any case it's mostly a management failure. What it can show however is that AI-first development requires more rigor, tests, and understanding of the codebase. In general, it requires you to move slower (ignoring the time spent typing), not faster. People tend focus too much on this typing the code part. But software doesn't get better by having more code, in fact I'd argue the oposite is true: The more there is, the larger the surface area for problems is, especially if most of said code has never been validated by a human.
In How AI impacts skill formation, J. Shen and A. Tamkin show that relying on AI hinders, stops, or even degrades human comprehension of the task at hand and knowledge of the code that solves it - the exact skills required to properly review generated code. In this situation, what would you do if someone reported a problem in your library? You could go through the tedious task of vading through the AI code trying to figure out what's going on - or - you could feed the issue back into the machine, hoping it will do it right this time.
If you adopt AI code as a library, that responsibility will eventually fall on you.
This website tries to report common signs of AI-first development. The hypothesis is that these correlate with a poor understanding of the codebase by the maintainers, which in turn is inversely related to their ability to improve or maintain the libraries quality over time, review code changes, respond to issues, and possibly also care long-term about their work.
Keep in mind that a repository with zero AI signals may still contain AI-written code. Not every tool leaves traces, and not every developer opts to commit their agent files. Conversely, a repository with many AI commits may be perfectly well-maintained by someone who reviews every line. Folks have also started to include agent files like this to deter others from using AI to contribute to their project.