IdeaWell

Project ideas from Hacker News discussions.

How much of F-Droid is LLM generated?

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

Theme 1: AI‑assisted coding is already common and often improves productivity or capability
- “I’ve basically stopped writing code by hand in 2026. The LLM writes better code than I do, much better.” – edg5000
- “My entire programming flow does not even include an editor… Writing good specifications is the key detail… you can absolutely generate high quality code if you know what you're doing.” – Zambyte
- “AI can help experienced engineers write better code in less time.” – lrvick

Theme 2: Concerns about AI‑generated code quality, maintainability, and over‑reliance
- “They produce pretty crappy code… they are very far from [writing good code] right now.” – jeltz
- “It’s extremely verbose, hard to read, often repeats code… review load went way up with AI.” – roblabla
- “If you don’t know how to use these tools, you’ll end up with crap. If you DO know how to use them, they are incredibly useful.” – LatencyKills

Theme 3: The debate over whether the origin of code (human vs AI) matters for end‑users or ethics
- “If they work, does it matter?” – alienbaby (repeated by several)
- “The burden of proof is on the new technology… I haven’t seen the efficiency benefits of AI trickle down to the programs I’m an end‑user of.” – rossy
- “For pre‑packaged code… it being AI or not is a problem only if the concerns we have about AI code… also apply to developer written code.” – alienbaby

🚀 Project Ideas

Generating project ideas…

LLMCodeAudit

Summary

  • Static analysis tool that flags code patterns commonly produced by LLMs (e.g., repetitive blocks, lack of abstraction, over‑reliance on boilerplate, co‑author tags, unusually large initial commits).
  • Core value proposition: Helps maintainers and reviewers quickly spot AI‑generated slop, reducing review overhead and improving long‑term maintainability.

Details

Key Value
Target Audience Open‑source maintainers, enterprise code reviewers, FDroid app curators
Core Feature Rule‑based + ML detector for LLM‑typical code smells, integrates with GitHub/GitLab CI
Tech Stack Python (ast, tree-sitter), scikit‑learn for heuristic model, GitHub Actions, optional WASM for editor plugins
Difficulty Medium
Monetization Revenue-ready: SaaS tiered pricing (free for public repos, $10/user/mo for private)

Notes

  • HN users complained about “vibe‑coded” apps and the difficulty of spotting AI‑generated code (“How much on F-Droid is LLM generated?”, “large init commits”, “co‑author” signals). This tool directly addresses that pain.
  • Could spark discussion on what constitutes acceptable AI assistance vs. harmful slop, and encourage better prompting practices.

FDroidAIscan

Summary

  • Web service that scans FDroid metadata and source repositories to estimate the likelihood that an app was vibe‑coded or heavily LLM‑assisted, using commit history, file‑size spikes, release frequency, and co‑author tags.
  • Core value proposition: Gives users and reviewers a transparent AI‑usage score to make informed decisions about trusting or promoting FDroid apps.

Details

Key Value
Target Audience FDroid users, app curators, security researchers, privacy‑conscious Android users
Core Feature Scoring dashboard + API that aggregates repo signals into an “AI‑likelihood” percentile per app
Tech Stack Go for scraping/analysis, PostgreSQL for storing metrics, React frontend, hosted on Fly.io or similar
Difficulty Medium
Monetization Hobby (open‑source, optional donations) – could later offer premium API for enterprises

Notes

  • Commenters expressed curiosity and concern: “How much on F-Droid is LLM generated?” and worries about low‑quality vibe‑coded apps flooding the store. FDroidAIscan gives them a concrete way to answer that.
  • Enables community discussion about thresholds, encourages developers to improve AI‑assisted practices, and could be integrated into FDroid’s review pipeline.

SpecCraft

Summary

  • Guided specification‑first assistant: developers write concise feature specs (in markdown or DSL), the tool uses an LLM to generate implementation scaffolding, then runs automated tests, suggests refactorings, and iterates until the spec is satisfied.
  • Core value proposition: Turns the vague “vibe coding” workflow into a repeatable, test‑driven process that maintains code quality while still leveraging AI speed.

Details

Key Value
Target Audience Solo developers, small teams, indie hackers who want to use LLMs without sacrificing maintainability
Core Feature Spec‑to‑code pipeline with built‑in test generation, linting, and iterative improvement loops
Tech Stack TypeScript/VS Code extension, Ollama or local LLM API, Jest/Vitest for testing, Prettier/ESLint for formatting
Difficulty High
Monetization Revenue-ready: $8/mo per user (team plans) – free tier for open‑spec projects

Notes

  • Several HN remarks highlighted that “writing good specifications is the key detail” and that they now spend time reviewing AI code rather than writing it. SpecCraft formalizes that workflow.
  • Provides a practical utility that can be demoed on HN, likely to attract interest from those experimenting with pi.dev/herdr/ollama setups.

Read Later