Claude for Deep Work: Projects, Memory & the 5-Day Setup Guide

This Claude setup guide covers what most users miss. Claude is the AI most people discover last — and use most. Here’s how to configure it properly so it actually works for deep work.

Table of Contents

1. What Makes Claude Different
2. Step 1 — Set Up Your First Project
3. Step 2 — Configure Memory
4. Step 3 — Introduce Yourself to Claude (Properly)
5. Step 4 — Claude Code for Non-Developers
6. Key Features Deep Dive
7. Real Workflow Examples
8. Power User Tips & Shortcuts
9. 30-Day Claude Challenge

What Makes Claude Different

There’s a moment most Claude users describe the same way: they paste a long, complex document — a contract, a strategy deck, a research report — and Claude responds as if it actually read it. Not just the first few pages. All of it.

That’s not a small thing. It changes what you can ask, how you work, and — if you’ve been using AI tools for a while — what you thought was possible.

Claude’s 200,000 token context window means it can hold roughly 150,000 words in its working memory simultaneously. A full-length novel. A year’s worth of meeting notes. Every email thread from a six-month project. Most AI tools give you a chapter. Claude gives you the whole book.

But context window alone doesn’t explain why people who try Claude tend to keep using it. The more significant difference is subtler: Claude reasons carefully rather than quickly. It’s more likely to say “I’m not certain” when it isn’t, more likely to identify what you actually need rather than what you literally asked for, and more likely to produce writing that sounds like a person rather than a template.

This Claude setup guide covers how to configure it so those qualities actually show up in your work — not just occasionally, but consistently.

Not sure how Claude compares to ChatGPT and Gemini? I break down all three in this side-by-side comparison →

Step 1 — Set Up Your First Project

Projects are Claude’s most powerful and most underused feature. A Project is a persistent workspace where Claude maintains context across multiple conversations — without you having to re-explain yourself every time.

How to create a Project:
Claude.ai → New Project → Name it for a specific area of your work

What Goes in a Project

Project Knowledge: Documents that define the project’s context — background materials, reference files, brand guidelines, strategy documents. Claude reads these at the start of every conversation within the Project.

Project Instructions: A system prompt that tells Claude its role within this Project, how to format responses, what to prioritize, and what to avoid. This is where you configure Claude’s behavior for this specific context.

Example Project: Content Writing

Project Knowledge:
- Upload: your brand voice guide, past articles, competitor examples

Project Instructions:
You are my writing partner for pickenough.com.
- Audience: knowledge workers who use AI in their daily work
- Tone: direct, slightly informal, no corporate-speak
- Format: use headers, short paragraphs, occasional bold for key points
- Always: suggest one internal link opportunity per article
- Never: use phrases like "In conclusion" or "It's worth noting"

Once configured, every conversation in this Project starts with Claude already knowing your publication, your audience, your tone, and your formatting preferences. You open a new chat and say “Draft an introduction for a post about AI decision-making” — and it does, in your voice, without setup.

Start with one Project for your highest-volume work. Once you’ve seen how much setup time this saves, you’ll build more.

Step 2 — Configure Memory

Claude’s memory works differently from ChatGPT’s — it’s more structured and requires slightly more intentional setup, but the results are more reliable.

Claude remembers information you tell it within Projects (persistent) and within conversations (session-only). Open a Project conversation and tell Claude:

Please remember the following about me and my work:
- My role: [your actual role and what you do]
- My goals for this project: [specific outcomes you're working toward]
- My working style: [how you think, what you value in AI responses]
- My audience: [who you're writing for or working with]
- Preferences: [response length, format, tone, what to skip]
- Things I don't want: [common AI behaviors that frustrate you]

Then test it: start a new conversation in the same Project and ask “What do you know about me and this project?” If the memory is working correctly, Claude should reflect back an accurate summary.

The difference this makes:
Without memory, you spend the first paragraph of every conversation re-establishing context. With it, you start from wherever you left off. For ongoing projects, this compounds significantly — the difference between an AI tool and an AI collaborator.

Step 3 — Introduce Yourself to Claude (Properly)

Beyond memory, there’s an underused technique that improves Claude’s responses immediately: writing a structured self-introduction and keeping it in your Project Knowledge.

This isn’t just “I’m a marketer who writes blog posts.” It’s a detailed briefing — the kind you’d give a skilled contractor on their first day.

Self-Introduction Template

About Me:
I'm [name], and I work as [role] at/for [context].
My main goal right now is [specific objective — be concrete].

My strengths: [2-3 things you're genuinely good at]
My weaknesses: [2-3 areas where you need the most help]
— Claude should compensate for these, not just agree with me

My communication style:
- I prefer [directness level]
- Response length: [short and punchy / detailed and thorough]
- Format: [bullet points / prose / tables / mixed]

What I don't want:
- [Specific behaviors that waste your time]
- [Common AI habits you want Claude to avoid]

When I'm wrong, I want Claude to:
[Tell me directly / ask a clarifying question / suggest an alternative]

This document lives in your Project Knowledge and runs silently in the background. You don’t reference it; Claude does. The result is a consistent working relationship rather than a blank slate every session.

Step 4 — Claude Code for Non-Developers

Claude Code is the feature most intimidating to non-developers and the one that most often surprises them once they try it. It’s a command-line tool that gives Claude direct access to your computer — the ability to read files, write code, execute scripts, and automate repetitive file operations.

You don’t need to know how to code. You tell Claude what you want in plain language, and it writes and runs the code to do it.

Accessible Use Cases for Non-Developers

File organization:
“Organize all the PDFs in my Downloads folder by month they were created, and rename them with the format [YYYY-MM-DD] [original name].”

Data processing:
“Take this CSV of sales data and generate a summary showing total revenue by category, month over month.”

Batch operations:
“Resize all images in this folder to 1200px wide and convert them to JPEG.”

Document conversion:
“Convert all these Word documents to PDFs and save them to a new folder called ‘Converted’.”

These are tasks that previously required either technical skills or expensive software. With Claude Code, you describe what you need, Claude writes and runs the script, and you review the result.

How to start:
Claude Code is available as a desktop application. Install it, open a project folder, and start with a simple, low-stakes task to build confidence with the tool before moving to more complex operations.

Key Features Deep Dive

200K Token Context Window
Claude can analyze your entire project history, read a book-length document in full, or maintain coherent reasoning across an extremely long conversation. For document-heavy work, this is a genuine capability difference — not a marginal improvement.

Writing Quality
Claude’s prose is consistently the most nuanced of the major AI tools. It’s less prone to over-structuring, less likely to use filler phrases, and better at matching a specific voice when given examples. For anything where the words themselves matter — proposals, essays, communications with real consequences — Claude produces output that requires less rewriting.

Honest Uncertainty
Claude is more likely than other AI tools to say “I’m not certain about this” when it isn’t. An AI that confidently provides wrong information is worse than one that flags its uncertainty. For high-stakes decisions where accuracy matters, Claude’s epistemic honesty is a practical advantage.

Cowork (Desktop Integration)
Claude’s desktop tool allows it to interact with your local files and applications — similar to Claude Code but with a more accessible interface for general file and task management. For non-developers who want to automate file operations without a command line, Cowork is the entry point.

Real Workflow Examples

Workflow 1: Long Document Analysis
I upload a 60-page strategy document to my Claude Project. I ask it to identify the top three strategic risks, note any internal contradictions, and summarize the recommended actions in under 200 words. Claude reads the full document and responds with specific references. What used to require two hours of careful reading takes 15 minutes.

Workflow 2: Writing in My Voice
My Project Knowledge includes several of my past articles and a detailed description of my writing style. When I ask Claude to draft a new post, it references these automatically — matching vocabulary, sentence rhythm, and structural approach. I still rewrite significantly, but the first draft is closer to my voice than generic AI output.

Workflow 3: Decision Support
Before making a significant decision, I describe the situation in full — context, constraints, options, and what I’m uncertain about. Claude identifies what I might be missing, suggests a framework for evaluation, and notes where my reasoning seems solid versus where it might be weak.

Workflow 4: Meeting Preparation with Full Context
My Project includes all relevant background for an ongoing client relationship — past proposals, meeting notes, email summaries, and our agreed scope. Before any client meeting, I open the Project and ask Claude to brief me: what did we commit to last time, what’s outstanding, and what questions should I expect?

Power User Tips & Shortcuts

Tip 1: Use Projects as a second brain, not just a chat window
The more you put into a Project — documents, notes, context — the more useful Claude becomes. Think of Projects as a living workspace that gets smarter as you add to it.

Tip 2: Give Claude examples, not just instructions
“Write in a professional tone” is vague. “Write like this example” — paste a paragraph you like — gives Claude something concrete to match. Examples almost always produce better results than adjectives.

Tip 3: Ask Claude to identify what you haven’t asked
At the end of a complex request, add: “What important considerations haven’t I asked about?” Claude’s ability to identify the question behind the question is genuinely useful for complex or high-stakes work.

Tip 4: Use long context for retrospective analysis
Upload all the notes, outputs, and reflections from a completed project and ask Claude to identify patterns, recurring issues, and lessons for future projects. With a large enough context window, it can see patterns you’d miss reviewing piece by piece.

Tip 5: Configure Claude’s pushback explicitly
In your Project Instructions, add: “If my request contains a logical flaw, point it out before proceeding.” By default, AI tools are agreeable. Making Claude a genuine thinking partner requires giving it explicit permission to push back.

30-Day Claude Challenge

Week 1 — Foundation

• Day 1: Create your first Project for your primary work area
• Day 2: Write your self-introduction and upload it to Project Knowledge
• Day 3: Configure Project Instructions for your specific context
• Day 4–7: Use Claude for one real task per day in your Project, note what changes

Week 2 — Context

• Upload all relevant background documents to your Project
• Test Claude’s long-context capability with a long document
• Try the “what haven’t I asked about?” technique on a current problem

Week 3 — Depth

• Build a second Project for a different area of your work
• Try Claude Code or Cowork for one file operation
• Use Claude for one decision where you’d normally just make the call intuitively

Week 4 — Evaluate

• Where did Claude’s depth make a meaningful difference?
• What tasks still required significant editing or correction?
• Where are you now more productive than you were before?

The Honest Comparison

Claude is the right choice when depth, accuracy, and sustained context matter more than speed or breadth. It’s not the fastest. It’s not the most versatile. And unlike Gemini, it doesn’t integrate natively with your email and calendar.

What it does better than any other AI tool available today: hold the full context of complex work and reason carefully within it.

For quick tasks, brainstorming, and creative variety, ChatGPT has the edge → For research synthesis and Google Workspace integration, Gemini’s integration changes the workflow →

But for the work that actually matters — long documents, strategic thinking, writing where the words need to be right — Claude is the tool I reach for. And once you’ve set up Projects properly, you’ll understand why the people who use it tend not to stop.

Still deciding which AI fits your work best? This comparison guide breaks down all 3 side by side →

There’s also a deeper question worth exploring: how does relying on AI tools change the way you think and make decisions? I explore that in this piece on the psychology of AI dependency →

And if you want prompts that work across Claude, ChatGPT, and Gemini for real work situations, the Work Prompt Pack has 45 tested prompts across four categories. $19.

Which Claude feature surprised you the most? Drop it in the comments.