A developer trained Claude Code on a library of 100 non-fiction books to discover thematic links, creating what they call 'trails.' The AI mined excerpts connecting ideas like startup deception to social psychology, linking Steve Jobs to Theranos. This moves beyond simple summarization, using LLMs to synthesize deeper insights across texts.

The system works by splitting books into 500-word chunks, indexing them by topic, and building a hierarchical tree for exploration. Claude uses CLI tools to search for co-occurring topics and generate novel connections. The author found the agent better at orchestrating this complex task than a hand-built pipeline, automating tedious revisions.

This project demonstrates AI agents can handle nuanced, creative workflows beyond coding. By treating Claude as a collaborative coworker, the author could focus on high-level design while the AI managed implementation. The approach opens new avenues for research and analysis, making revision cheap and ambitious projects more feasible.