Documenting your open source projects for machines
summary
Claims that LLM-driven agents read and prefer machine-friendly docs, with concrete steps and tools: publish markdown in wheels, LLMs.ext sitemap pages, use sphinx-llm or mkdocs-llmstxt and ship docs alongside the package so local agents avoid stale web copies. Gives examples of agent behavior (defaults to pandas, reimplements slow groupby), recommends type hints, clearer error messages and skills files, and shows how to test by watching an agent import read run and fix code.
AI-written summary · about
chapters (15)
- 0:00Documentation for AI agents
- 2:07How agents learn libraries
- 6:45Undocumented libraries challenges
- 9:13Agents consume documentation
- 11:19Progressive disclosure
- 13:46Machine-readable docs
- 16:57Markdown docs tools
- 19:43Docs in Python packages
- 22:09Diataxis framework
- 25:15Agent-focused documentation
- 29:05Human vs AI docs balance
- 31:10Libraries for agent usability
- 32:57Agent-friendly code practices
- 34:41Testing docs with agents
- 36:06Conclusion and key takeaways
related talks
49mNext-Gen Documentation with AI- 49mNext-Gen Documentation with AI
30mBuild an AI knowledge fabric for your organisation
19mLLM codegen fails and how to stop 'em
41mBuilding LLM Agents – Challenges and Learnings
01h 28mBuilding LLM Agents Made Simple a
38mNobody Pages the LLM: Engineering Rigour for Vibe Coding
23mLLM Agents in Production: Why Go is the Missing Piece