Real example · Study, summaries & research
Found hidden links across 100 books with Claude Code
What they did
Pieter Maes indexed 100 non-fiction books and gave Claude Code command-line search tools so it could chain short excerpts from different books around one idea, which he calls trails, instead of using a fixed multi-step pipeline.
How it went
Topic extraction over the whole library cost roughly £10 (about $13) using a cheap Gemini model, and the project site shows 44 trails. He found the agent-with-tools approach worked better than his hand-built pipeline.
Worth knowing
Many extracted topics are low value, the agent's ideas kept drifting to the same few themes, and prompt tuning was only one rough pass. Time spent and the total number of trails tried are not reported.
Try it yourself with Claude Code
Replace anything in [brackets] with your details.
I have [number] documents in [folder] and a search tool [describe it]. Find 5 trails: each is a sequence of 4 to 6 short excerpts from different documents that build on one idea, with a one-line note on how each excerpt connects to the next.
Avoid these themes I have already seen: [list]. Quote the exact source and location for every excerpt so I can check it. Read the original
Reading across books with Claude Code ↗
pieterma.es · · Personal post
Attributed to the source linked below; not our own test.
Share Share on X Share on WhatsApp
