This is the first step. Skip it, and everything downstream is generic.
The brain is the thin layer Codex actually reads, distilled out of the dump.
Week 1 is an identity, not a feature list.
By the end of the week, Codex knows all three.
The four-step method, run on each person's own real material, not a blank page.
about-me, voice-reference, and working-preferences are slow. current-context is fast.
Previewed now, enforced from Week 4 once the recording layer lands.
We're tool-agnostic. The architecture works on Claude, Codex, Gemini. We default to Claude because it's the best fit for most engagements. For specific principals or workflows we pick the right tool for that slot. The architecture stays the same.
Short sessions only work if the chain behind them holds.
One 45 to 60 minute session, landing alongside this week. No keyboard exercises.
Rob's rule, standing for the life of this course.