How much of F-Droid is LLM generated?

F-Droid, a repository of free and open-source software, has been the subject of speculation regarding the extent of AI-generated content within its offerings. A reporter sought to determine the proportion of apps within F-Droid that were likely created using large language models (LLMs). The investigation focused on a batch of 102 apps updated to F-Droid on September 12, 2026.

To gauge the likelihood of AI involvement, the reporter analyzed several factors including recent commit history, project branding, commit messages, and the presence of AI-related tools in the repository. Apps exhibiting strong indications of LLM influence, such as automated README generation, automated code reviews, and excessive use of emojis, were categorized as "mostly AI."

Among the 102 apps, several were deemed to have a high probability of AI-generated content. For example, the "Amber" app, with all recent commits attributed to LLMs and PRs accepted from AI agents, was classified as "mostly AI." Similarly, "Aria for Misskey" and "Baly Groceries Tracker" displayed AI-related code and commit characteristics, leading to their categorization as "mostly AI" as well.

However, not all apps fell into the "mostly AI" category. Some, like "Atmo Engine," showed signs of AI influence but lacked strong indicators, resulting in a more neutral classification. The reporter acknowledged the limitations of their assessment, noting that the absence of a definitive "slop detector" meant that their findings were based on superficial analysis and may contain errors.

In conclusion, the reporter determined that a significant portion of F-Droid's offerings, potentially as high as 50%, may have been AI-generated. While the precise percentage remains uncertain due to the challenges in accurately detecting AI involvement, the trend suggests a growing reliance on LLMs in the development of open-source software.