If Claude retained nothing between sessions, what would you change about how you work? That question reframes how you think about projects — what good instructions actually contain, what good files look like, and where everything else needs to live instead.

The difference between Claude giving you generic output and Claude working inside your reality is context. Four layers of it — user preferences, memory, a voice profile, and context documents — each with a hands-on exercise you can do right now.

Everything I know about working with Claude well, I learned from working with Claude Code badly. After a month of duplicate work, lost decisions, and re-explaining things I'd already unpacked, I developed a three-part session protocol — orient, do bounded work, close the loop — that quietly compounds the quality of every conversation you have with AI.