Skip to content

AI_DEVELOPMENT.md

Latest commit

 

History

History
79 lines (54 loc) · 3.94 KB

AI_DEVELOPMENT.md

File metadata and controls

79 lines (54 loc) · 3.94 KB

🤖 AI-Assisted Development

Philosophy

libdrmtap is built using AI coding agents as first-class development tools. We believe this represents the future of open-source development, and we're transparent about it.

We don't hide AI usage — we document it, celebrate it, and encourage it.

Why AI Agents?

The DRM/KMS screen capture problem requires deep knowledge across multiple domains:

  • Linux kernel internals (DRM subsystem, ioctls, DMA-BUF)
  • GPU driver specifics (Intel, AMD, Nvidia — each completely different)
  • Security models (CAP_SYS_ADMIN, DRM Master, Unix sockets, SCM_RIGHTS)
  • Pixel format handling (tiling, deswizzle, HDR tone mapping)
  • Cross-referencing dozens of existing projects, issues, and kernel docs

A single developer researching all of this manually would take weeks. With AI agents, the research phase — analyzing 11 projects, reading 25+ GitHub issues, studying 3 reference codebases line by line — was completed in hours.

Tools Used

Antigravity (Google DeepMind)

  • Role: Primary agentic coding assistant
  • Used for: Codebase analysis, GitHub issue research, architecture design, documentation writing, code generation
  • Why: Full agent capabilities — can browse GitHub, read source code, analyze issues, and produce structured technical documents

Gemini (Google DeepMind)

  • Role: Research and analysis engine
  • Used for: Web searches, cross-referencing kernel documentation, understanding DRM APIs
  • Why: Strong reasoning over technical documentation and API specifications

Claude (Anthropic)

  • Role: Code generation and technical reasoning
  • Used for: Code generation, API design review, documentation structure
  • Why: Excellent at generating clean, well-structured C code and technical writing

What AI Did

Research Phase (documented in docs/research/)

  1. Landscape analysis — Found and analyzed 11 DRM/KMS projects on GitHub
  2. RustDesk deep dive — Traced 4 years of Wayland issues, PRs, and discussions
  3. Source code analysis — Read kmsvnc/drm.c (~800 lines), kms-screenshot.c (~1500 lines), FFmpeg kmsgrab.c line by line
  4. Issue autopsy — Read all 25 kmsvnc issues, categorized patterns, extracted gotchas
  5. API design — Compared 3 implementations, proposed unified API
  6. Architecture — Designed the helper binary pattern based on gpu-screen-recorder's approach

Implementation Phase (ongoing)

  • Guided by the research findings
  • Every architectural decision traceable to a specific finding in the research docs

What AI Did NOT Do

  • Make final decisions — A human developer reviewed and approved all designs
  • Test on real hardware — AI can't access GPU hardware
  • Replace understanding — The human developer understands every line of the research and code
  • Work unsupervised — Every step was reviewed, questioned, and refined through human-AI dialogue

How to Contribute with AI

If you use AI tools (Copilot, ChatGPT, Claude, Gemini, or others) to help with your contributions:

  1. ✅ Go for it — We encourage AI-assisted contributions
  2. ✅ Understand your code — Don't submit code you can't explain
  3. ✅ Test it — AI-generated code needs the same testing as human-written code
  4. ✅ Credit the tool — Optionally mention in your PR if AI was used (not required)
  5. ❌ Don't auto-generate and submit — Review, understand, then submit

The Thesis

A single developer with AI agents can produce research and code that would traditionally require a team of specialists.

This project is living proof of that thesis. The entire research corpus (7 documents, ~5000 lines of technical analysis) was produced by one human + AI agents in a single session.

We hope this inspires other open-source developers to embrace AI as a tool — not a replacement, but an amplifier of human capability.

"The best tool is the one that makes you more effective without making you less thoughtful."