10,497 prompts. 68 days. 53 projects.
What does a body of work look like when you're building with AI every day?
I ran an analysis on my Claude Code conversation history. Not for optimization. For archaeology.
Every prompt I've written since November 2025—extracted, timestamped, clustered by project. The raw material of building with AI.
Three questions drove this:
Each square is a day. Click to see the interleave pattern.
Three acts emerged from the data.
Nov 25 – Dec 16
"What can I do with this thing?"
5-8 projects running simultaneously. Nothing ships yet. Testing boundaries.
Dec 17 – Jan 6
"Build the core asset"
One project dominates. Christmas Day: 299 prompts on Source Library.
Jan 6 – Feb 4
"Ship products for users"
Intensity spike. Feb 2: 793 prompts. The experiments became products.
The real signature isn't parallel projects. It's multiplexed attention.
Not time-blocking. Not "morning on X, afternoon on Y." Constant interleaving at sub-minute intervals. Like a DJ mixing between tracks.
Each row shows how attention flows between projects throughout a single day. Click any day to see details below.
Each spike = a context switch. The height shows how many prompts before switching again.
When does the work happen?
Data extracted from ~/.claude/projects and ~/.claude/history.jsonl using custom Python scripts.
Built with D3.js for visualizations and Python for data extraction.