Move From Claude Chat to Cowork and Code
A two-page setup sheet plus the full playbook for project folders, instructions, memory, workflows, and reusable prompts.
Cheatsheet (condensed version)
Claude: Beginner to Pro
The complete setup cheat sheet. Everything from the video on two pages.
The 3 Levels
- Level 1: Chat. You type, Claude responds. You do all the work.
- Level 2: Cowork. Your assistant. Handles daily tasks on your desktop.
- Level 3: Claude Code. Your engineer. Builds tools, systems, automations.
Your Project Folder
Every project gets its own folder. Claude reads these files for context.
/Your-Project/ INSTRUCTIONS.md <- Employee handbook MEMORY.md <- Long-term brain (the hack) CONTEXT-strategy.md <- Reference docs for the job /outputs/ <- Where work goes
INSTRUCTIONS.md
This is your agent's employee handbook. Five sections:
- Who you are: One line. "You are my [role] for [project]."
- What you do: List the actual tasks. Be specific, not vague.
- Rules: Tone, fact-checking, when to ask questions vs guess.
- Good output: Describe ideal output. Length, format, style.
- Memory rule: "Read MEMORY.md at start. Update it at end."
Template (copy this):
# INSTRUCTIONS ## Who You Are You are my [role] for [project/company]. ## What You Do - [Specific task 1] - [Specific task 2] - [Specific task 3] ## Rules - Match my tone: [describe your voice] - Never invent stats - flag anything needing a source - If unclear, ask ONE question before starting - When you learn something new, add it to MEMORY.md ## What Good Output Looks Like - [Your format/length/style preferences] ## Memory Start of every session: read MEMORY.md End of every session: update MEMORY.md with new learnings
The Memory Hack
Claude doesn't remember between sessions. MEMORY.md fixes that.
- Add this to your INSTRUCTIONS.md: "Read MEMORY.md at the start of every session.
Cheat Sheet - Page 2
AI Edge
MEMORY.md Template
# MEMORY ## Preferences - [Format preferences, tone, tools, length] ## Corrections - [Mistakes Claude made + what the right answer was] ## Patterns - [Recurring tasks, common workflows, frequent requests] ## Decisions - [Choices made together that shouldn't be revisited] Don't edit this manually. Let Claude maintain it. It grows every session.
Context Files - What to Include
Only include docs relevant to the agent's job. Name them clearly.
- Content creator: Content strategy, brand voice, past scripts
- Business owner: Company overview, SOPs, product docs
- Freelancer: Client briefs, portfolio summary, rate card
- Developer: Architecture docs, coding standards, API specs Think: what would I hand a new employee on their first day?
The Pro Loop
This is how Cowork and Claude Code work together:
- Give Cowork a real task from your daily work.
- Notice when you keep giving the same type of task.
- That's your signal. Go to Claude Code.
- Code builds a system that automates that task permanently.
- Cowork's job shifts - it reviews instead of creates.
- Repeat. Every loop = less manual work.
The Compounding Timeline
- Week 1: You're handing off individual tasks.
- Week 4: You've built systems that handle whole categories of work.
- Week 8: The agents are compounding while you sleep.
20-Minute Quick Start
[ ] Create your project folder. [ ] Write INSTRUCTIONS.md using the template on page 1. [ ] Create MEMORY.md with the four sections (preferences, corrections, patterns, decisions). [ ] Add one or two context files relevant to your work. [ ] Open Cowork and give it a real task. Not a test - something you'd actually do today.