Guide on working effectively with AI by organizing context, encoding preferences as configuration, and building skills that compound over time.

The author proposes treating AI collaboration like onboarding a new teammate: organize code and knowledge into clean directory structures (e.g., ~/src for code, ~/vault for knowledge work), create per-project INDEX.md files with annotated links to relevant docs and channels, and treat per-project CLAUDE.md as an onboarding document with glossaries, code names, and suggested reading order.

Preferences and workflows should be encoded in configuration files rather than stated repeatedly: write behavioral contracts in ~/.claude/CLAUDE.md (how direct to be, when to push back, how to handle mistakes), lazy-load longer guides by topic (writing, evals, dashboards), and define skills (reusable markdown workflows) for tasks done ≥ once a week, bootstrapping them by doing the task interactively first and then asking the model to formalize it.

Every finished artifact—code, docs, analysis, decisions—should become context for the next session, with corrections and feedback logged in session transcripts; this approach avoids hallucination and reduces future errors by letting the model build on prior work and refine skills through iterative feedback cycles rather than direct file edits.