Notes
How I Work With Claude Code
Figma to Claude Code to TestFlight, an awaiting-review step, a decision log, and a did/decided split.
Losing to Blender
My AI agent lost a speed test to Blender, 543 s to 169 s. I changed the pitch, not the benchmark.
What I Chose Not to Automate
Kivo's reships, Lingisa's review step and Orchid Finance's bank APIs: where a person stayed in the loop.