Claude Projects Playbook

How much of your AI session time is spent re-explaining what you already explained last time?

Claude Projects eliminate that loop. Each project is a persistent workspace where Claude holds your context — business rules, client details, templates, SOPs — and carries it into every conversation without prompting.

What Projects Are

A Project is a named workspace with two layers:

Layer	What it holds	Effect
Instructions	Role, tone, goals, constraints	Claude behaves consistently without prompting
Files	Documents, context, templates, SOPs	Claude references these in every chat automatically

All chats inside a project share the same context. History accumulates. The assistant gets sharper the more you use it.

Setup Pattern

1. Name It

Name the project after its purpose — client name, function, role. Specificity here reduces drift. "Acme AI Consulting" beats "Client Work."

2. Write Instructions

Instructions define what this assistant is. Not what it can do — what it is inside this project:

• Role: what function it performs
• Scope: what tasks it handles
• Style: how it communicates
• Constraints: what it never does

If you don't know what to write, ask Claude to interview you. Answer its questions about your goals, then have it draft the instructions from your answers.

3. Load the Files

Files are the knowledge base. Good starting files:

File	What goes in it
Business context	How your operation works, what you do, how you position
Client context	Who they are, what they need, their pain and constraints
SOPs	Step-by-step processes for recurring work
Templates	Standard questions, email formats, audit frameworks

Upload documents, paste text, or link to Google Drive. Title each file clearly — Claude references files by name when reasoning about what to pull.

Research shortcut: Use Claude's deep research mode to build the client context file automatically. Give it the company URL. It searches and compiles a business brief in under five minutes.

Core Loop

Meeting → Paste notes → Ask Claude → Draft output → Update project files

Every loop adds to the project. Over time the assistant knows the full history — who said what, what was decided, what comes next.

Meeting Follow-Up

Paste meeting notes. Ask: "Based on these notes and your knowledge of [client], create an action plan, surface key decisions, and draft a follow-up email."

Claude produces the action plan, the email, and flags dependencies — budget questions, capacity constraints, open items — using context already in the project. What used to take an hour of note-scanning and draft-revising takes minutes.

Template Customization

Any template becomes a starting point, not a blank page. Add your standard questions, audit framework, or checklist to the project files. Then ask: "Tailor this for [specific client/industry] based on what you know."

The output is specific — not generic advice — because the project holds the context that makes it specific.

Repetitive Work

Every recurring task that has a pattern is a candidate for a project SOP. Add the SOP to project files. From then on, just trigger it: "Run the post-meeting SOP for today's notes." Claude executes the same steps every time without re-prompting.

Why Projects Beat Chats

Chat	Project
Re-explain every session	Context loads automatically
Scattered history	All conversations in one place
Generic outputs	Specific to accumulated context
Prompts from scratch	Prompts get shorter over time
Breaks when context window fills	Files persist across all sessions

The pattern compounds. Early sessions are efficient. Later sessions are nearly frictionless — Claude already knows the client, the workflow, and the standard.

Build Order

Start lean. Add complexity only when you feel the friction of what is missing:

1. Session 1: Instructions + business context file
2. Session 2–3: Client context file + first template
3. Ongoing: Add SOPs after you've done a task manually twice

Over-building upfront produces a project nobody maintains. Build from use.

Questions

What recurring task in your current workflow requires the most context-loading before Claude can be useful — and what file would eliminate that load if it lived in a project?

• Which of your clients or projects would benefit most from having Claude remember the full history, not just the last chat?
• If Claude already knew everything in your project files, what would you stop re-explaining in every session?
• At what point does a project's accumulated context become a liability rather than an asset — and how do you prune it?