I've had 2,700 conversations with AI...
Problem: Over 2,700 Markdown Documents: The Result of an Obsession and Compulsion with Notes
I exported my conversations from AI services like ChatGPT, Grok, and Claude.
I asked an AI how to turn the conversation.json file into Markdown using a Python script. After converting everything, I ended up with more than 2,700 documents.
There must be important notes hidden somewhere in all those files, and I’m not sure what to do.
The easiest solution would be to delete everything and start over — but I can’t do that.
At first, I planned to use an LLM to summarize and export all the documents. But then I realized it would be better if I delete the files myself.
In the end, what I remember and understand stays with me only if I do the work myself.
Let's solve the problem using Claude Code.
I asked Claude, and it told me that in Claude Code CLI, I can use the -p flag to send a file through a pipe and get the result right away.
This is the main command:
cat "$(find . -name '*.md' | head -1)" | claude -p "Extract only the key points from this document as bullet points. Output only bullets, no extra explanation."
(For reference, I was using Claude Max $200 with Claude 4.6 Opus and 4.6 Sonnet at the time.)
The focus of this article is not how it was made. Anyone can use AI to create a summary script or Python code.
I experimented with different methods and even thought about documenting them, but that wasn’t the main message.
What matters most is doing the thinking myself. I only truly understand something when I work through it on my own.