Analyzing the AI-Saturation of F-Droid

Date: September 15, 2026

📑 Table of Contents

Introduction
The Core Issue
Personal Biases
The Methodology
The App Audit
Final Thoughts

Introduction

I have a deep affection for F-Droid. I admire the philosophy it champions and the digital autonomy it grants its user base. As a maintainer of FOSS software myself, my experiences with the team have been exclusively positive.

In fact, they might be too lenient—I've lost count of how many times my repository builds failed simply because I forgot to commit my changes before triggering the build process 😅.

The Core Issue {#the-core-issue}

Occasionally, I use F-Droid as a discovery tool, browsing for a utility that solves a latent problem or searching for a superior alternative to a current tool. During one such session, I encountered an app with a glaringly artificial, AI-generated icon.

While I won't publicly shame the developer, it sparked a question: How much of the FOSS ecosystem is now being vibe-coded?

As a student who codes for pleasure and remains "unspoiled" by professional corporate labor, my perspective on the industry is shaped largely by Reddit threads and sensationalist YouTube videos. The discourse usually falls into two extremes:

LLMs are ~~divine incarnations of intelligence~~.
LLMs are ~~merely fancy autocomplete tools~~.

Both perspectives are reductive. I wanted to know the actual probability that a FOSS app was slapped together by a random person in a single afternoon using an LLM.

Since raw text doesn't provide enough metadata for a precise audit, I had to look for "smells." Much like a specific punctuation choice can signal a certain writing style, LLM-generated repositories leave footprints. The primary draw of AI is the ability to be lazy, and that laziness permeates the project's hygiene.

The "Vibe-Coder" Red Flags:

Allowing the LLM to review its own pull requests.
Granting the AI direct write access to the repo to bypass manual commits.
A general lack of manual curation in the project's history.

Personal Biases {#personal-biases}

To avoid the typical "circlejerk" and provide transparency, I'll state my position on Large Language Models.

I recognize that LLMs are powerful. By 2026, the ability to "one-shot" a medium-sized piece of software or a game in thirty minutes is objectively stunning, even if the quality is often mediocre. I also support the use of AI to enhance security and performance; the Linux kernel's development has proven that LLMs can identify and patch complex bugs.

However, I fundamentally dislike what they've done to the craft of programming.

Educational Erosion: The joy of a side project is diminished when you realize an LLM could have done it in 25% of the time.
Cognitive Atrophy: We are seeing a decline in deep problem-solving skills.
Misinformation: The capacity for these models to generate "confident" falsehoods is terrifying.
Environmental Cost: The massive energy consumption required to sustain these models is a climate disaster we are largely ignoring.

The Methodology {#the-methodology}

Because there is no perfect "slop detector," I developed a qualitative three-tier classification system based on repository aesthetics and commit history.

Classification Matrix

Tier	Label	Description	Criteria
Tier 1	`Mostly AI`	High "slop" scent; >50% AI authored.	Use of agentic coding harnesses or AI-driven commit loops.
Tier 2	`Mixed/Human`	Primarily human-led.	Occasional AI assistance in commits, but maintains human oversight.
Tier 3	`No AI Signs`	Clean/Human.	No suspicious patterns or an explicit "No AI" policy.

The Logic Flow

To quantify the "slop factor" ( $S$ ), one might imagine a formula like: $S = \frac{\text{AI Commits} \times \text{Lack of Documentation}}{\text{Manual Code Reviews}}$

The App Audit {#the-app-audit}

I reviewed a massive array of applications. My assessment was superficial, based on recent commits and project descriptions.

📋 Apps Evaluated

Visual Evidence of "Vibe-Coding"

Below are some of the icons encountered during the research:

Aria for Misskey Atmo Engine Aves Libre Baly Groceries Tracker Bati: Fitness RPG BeatBridge: Bluetooth Music Binary Eye

Final Thoughts {#final-thoughts}

Throughout this process, I noted several interesting patterns. I looked into specific contributors like brandonp2412 and explored the prevalence of projects hosted on Codeberg. I also kept an eye out for "Don't tread on me" style apps.

To summarize the technical findings in a structured format:

{
  "audit_summary": {
    "platform": "F-Droid",
    "year": 2026,
    "primary_metric": "Repo Aesthetics",
    "findings": "Variable levels of AI integration",
    "sentiment": "Concerned but hopeful"
  }
}

Despite the rise of the "vibe-coder," my core sentiment remains unchanged: I love F-Droid.