Claude Cowork: 3-Level Hierarchy Builds AI Second Brain
Turn Claude into a persistent AI coworker using CLAUDE.md instruction files and memory.md for a 3-level hierarchy (root, workstations, projects) that handles emails, finances, newsletters, and projects without burning rate limits.
CLAUDE.md and Memory.md Enable Persistent Context
CLAUDE.md acts as the master instruction manual governing Claude Cowork behavior, loaded at every session start, while memory.md stores persistent details like active projects and recalled facts. Key rules in root CLAUDE.md include: "At the start of every session, read memory.md before responding" and "When I say 'remember this', write to memory.md." This surfaces implied context—writing style, projects, preferences—automatically, reducing manual repetition and improving outputs. Voice principles.md, extracted from 30 Gmail emails or samples via prompt templates, captures tone (e.g., "warm, direct, professional without being stiff") and evolves to 150+ lines. Routing map in CLAUDE.md directs tasks to workstations (e.g., copywriting frameworks to specific files). Resources folder holds referenced files loaded only when needed, keeping root CLAUDE.md under 300 lines to minimize token usage.
Active projects section in memory.md lists ongoing work (e.g., workshop outline, newsletter); tell Cowork "add this project" to update. Session audit skill (/session-audit) scans chats for unsaved principles, appending to memory.md or CLAUDE.md.
3-Level Hierarchy Stacks Rules for Specialized Workflows
Root level (Level 0) applies universally like a constitution. Workstations (Level 1) add domain-specific rules stacking atop root: universal ones like Email HQ (cross-life tasks) analyze 4 weeks of sent emails for greetings/signoffs, inbox-zero workflow (2-minute rule, labels, archive/snooze logic); dedicated ones like Personal Finances process 12 months of credit card statements into Excel trackers (tabs: Transactions, Yearly/Monthly Summary, Category Taxonomy), learning corrections (e.g., Canva as "software/subscription" not "freelancer"). Each workstation auto-creates its own CLAUDE.md, memory.md, resources folder via prompts.
Projects (Level 2, under workstations) mirror this for single initiatives (e.g., mortgage refinance under Housing, trips under Travel), inheriting stacked rules. Start with 2-3 workstations; expand as needs arise. Obsidian previews markdown files readably; folder is single source of truth for all docs.
Use Cases and Token Optimization Deliver Production Results
Cowork routes screenshots to files, drafts follow-ups by pulling calendar/transcripts and referencing threads, creates Notion projects matching conventions (properties, sections, notes). Examples: finalize newsletter in user's voice linking Notion drafts; review expenses ($1,000+ on Bumble); process statements.
Optimize tokens: (1) Root CLAUDE.md <300 lines, reference external files; (2) No rule duplication across files; (3) Default to Sonnet model (1/5 Opus cost) unless 3+ interdependent steps. Pro tips: Star workspace for default load; download MD files properly; use Gmail connectors or samples for voice/email analysis; end sessions with /session-audit.
Download starter templates (CLAUDE.md, memory.md, voice principles.md, prompts for workstations) and free Cowork Toolkit for pre-built systems, skipping trial-and-error.