How much of F-Droid is LLM generated?

It is now easier to discover apps with AI-generated code on F-Droid.

A student programmer investigated the prevalence of AI-generated code within the F-Droid app repository, examining a batch of 102 app updates from September 12, 2026. The analysis focused on identifying "AI smells" in code, commit messages, and project documentation, such as automated README generation, AI-assisted code reviews, and direct LLM code contributions. The researcher categorized apps into three tiers: "Mostly AI" (>50% AI-authored code, including agentic infrastructure), "Hard to say / Mostly human / Other" (<50% AI-authored code with occasional LLM commits), and "No signs of AI" (no suspicious activity or strict LLM policies). Specific apps like Amber, Balance, Baly Groceries Tracker, and BayesianBahn were flagged as "Mostly AI" due to clear indications of AI involvement in their development, while others like Aria for Misskey and Atmo Engine were classified as having "No signs of AI" despite minor peculiarities. The researcher acknowledges the limitations of this superficial assessment, noting that definitive detection of AI-generated code is not possible and errors may occur.

The methodology, while informal, highlights a growing challenge for open-source repositories: distinguishing human-authored code from AI-generated content. The criteria used—examining commit messages, README files, and the overall project presentation—suggests that a significant portion of new development might be influenced by LLMs, potentially altering the landscape of software contribution and maintenance. This trend could impact the perceived authenticity and trustworthiness of open-source projects, as the effort and skill involved in development become less transparent.

The findings suggest that the allure of LLMs for rapid development is leading some maintainers to delegate substantial parts of the coding process to AI. This raises questions about the long-term sustainability of projects heavily reliant on AI-generated code, particularly concerning maintenance, security, and the originality of innovations. As LLM integration becomes more seamless, distinguishing genuine human-driven creativity from AI-assisted output will become a critical concern for users and the open-source community alike.