Claude Chat, Cowork, and Code: The Setup Most People Never Do
With templates, the context files and configuration that make each one actually work and know you.
This post is going to be all about configuring Claude Chat, Claude Cowork and Claude Code (desktop).
Most people solve the first one without only touching on it a little bit and don’t use those tools to their full potential.
And I get it, it takes time and can get confusing. But if you want to get the most out of these tools, you need to configure them properly.
They read a comparison guide, pick a tool, open it — and use it with zero setup. No context files. No instructions. No configuration. Claude starts every Chat session not knowing who they are. Cowork forgets their preferences the moment they close it. Code reads their files but not their working style.
The tools aren’t bad out of the box. They’re just not yours...yet.
I’ve spent the last few months running all three on real work — newsletter production, product research, writing, coding and building. I learned the hard way which settings matter and which ones you can ignore. This guide is the hour of setup I wish someone had handed me before I wasted time figuring it out session by session.
By the end of this post, you’ll have:
- Chat configured with a context file that loads every session
- Cowork set up with global instructions, one plugin, and one live connector
- Claude Code (desktop) running with a context file and your first custom skill
- A handoff pattern between all three so you stop forcing one tool to do another’s job
Block an hour this weekend. Let’s get into it.
How Claude Chat, Cowork, and Code Handle Memory Differently
Before the setup steps, one reframe.
Every comparison guide focuses on ease versus power. That’s fine, but it’s not the frame that helps you configure anything. The frame that matters for setup is memory and file access.
Chat has no memory between sessions. You bring everything to it - your context, your files, your preferences. Every session starts from zero.
Cowork has session-level memory. You point it at a folder and it reads everything inside. It can write files back to that folder too. But when you close the session, it forgets your preferences unless you’ve written them down somewhere it can find.
Claude Code (desktop) has persistent memory via context files - documents you write once that Claude reads at the start of every session. It reads your project files, understands your setup, and picks up where you left off.
The setup investment scales with the depth of access. Ten minutes for Chat. Thirty for Cowork. Forty-five for Code. Each one pays back every session after the first.
Setting up Chat the right way
Chat is where most people stop. It works, so they never touch the settings.
That’s leaving most of the value on the table. A Chat session with no configuration is Claude at maybe 40% capacity. Not because the model is limited, but because it doesn’t know anything about you.
How to Configure Claude Projects for Persistent Context
Let’s talk about Claude Projects. A Project is a persistent workspace that loads the same context every time you open it. You don’t re-explain yourself. You don’t paste your preferences into every prompt. You just start working.
Most people create a Project and leave it empty.
Here’s what to actually put inside one. 3 files, written once, updated occasionally:
File 1: context.md
Who you are, what you’re working on, what good looks like.
# Context
## Who I am
[2-3 sentences. Your role, what you build or create, who you serve.]
## What I’m working on
[Your current main project. One paragraph. What it is, where it stands, what matters right now.]
## What success looks like
[How you measure whether something you make is good. Be specific. Not “high quality” — what quality actually means for your work.]File 2: voice.md
For example, in my case, it’s a newsletter. So for me it would be all about how I write a it. This is the file that stops Claude from sounding like everyone else.
# Voice
## How I sound
[Describe your writing in 3-4 sentences as if explaining it to a new colleague. Direct? Casual? Opinionated?]
## Words I never use
- “Delve”, “tapestry”, “leverage” as a verb
- “Robust”, “seamless”, “cutting-edge”
- “Utilize” — just say “use”
- [Add your own. The more specific, the better.]
## What sounds wrong
[Patterns you keep correcting. Phrases that feel off when you see them. This is the most useful section — and most people leave it blank.]
## What I believe
[Your actual opinions about your topic. What you think most people get wrong. Your contrarian takes. This prevents generic output more than any style rule.]Most people stare at that template and write something vague like “professional but friendly.” That tells Claude nothing.
Here’s a better approach: interview yourself.
Record yourself answering specific questions about your identity, your beliefs, what makes you uncomfortable, and how you actually work. You can talk through it or type. Don’t edit yourself. The raw, slightly messy version of how you think is more useful than a carefully crafted description.
I built a Voice Profile with interview questions for exactly this (and for myself too). Five versions - one general, plus tailored versions for writers, marketers, developers, and founders. If you have a different role that I didn’t include, simply download one of the versions and ask Claude to create a tailored version for you. Once you answer the questions, compress the highlights into your “voice.md”.
The voice file you get from a 30-minute interview is ten times more useful than anything you’d write from a blank page.
Download the Voice Profile here (free)
File 3: notes.md
Running decisions and things to remember across sessions.
# Standing Notes
## Decisions made
[Things you’ve already decided. So Claude doesn’t re-suggest them every session.]
## Open questions
[Things you’re still figuring out. So Claude knows what’s still live.]
## Recent feedback
[What’s been working. What hasn’t. Updated periodically.]Load these three files into a Project (you can simply drag them in). Open a new session. Type this:
Read everything in this Project and tell me what you know about me, how I work, and what I’m building. Be specific. If anything is missing or unclear, flag it.
If the answer reflects your context accurately, you’re configured. If something’s wrong, fix the file - not the prompt. The file is the single source of truth.
Extended thinking: when it matters and when it doesn’t
One setting worth knowing. Extended thinking tells Claude to reason through a problem step-by-step before answering. It makes a real difference for: analyzing complex documents, weighing tradeoffs, planning multi-part projects. It’s overkill for: writing drafts, answering quick questions, generating ideas.
It uses more tokens. For the right task, the quality difference is obvious. For everything else, standard thinking is fine.
Your first action: Create a Project. Add the three files above. Run the calibration prompt. If the response sounds like you, move on. If not, the problem is in the files - tighten them and run it again.
That’s Chat. The next two sections - Cowork and Claude Code (desktop) - go deeper. Most of the configuration leverage is there.
We will go into the full Cowork setup (global instructions, folder strategy, plugins, and connectors), the Claude Code desktop setup (context files, custom skills, and the retro loop that makes it compound), the handoff workflow between all three, role-based configurations, and the full setup walkthrough.
Setting up Cowork the right way
Cowork launched in January 2026 and has more configuration surface than any other Claude interface. Most people use maybe 20% of it because the defaults are good enough to feel fine - but not good enough to feel like yours.
There are five layers to a proper Cowork setup. They build on each other. Do them in order.
Layer 1: The folder and context file strategy
The first thing Cowork asks you to do is select a folder. Most people pick whatever folder is most convenient. That’s the wrong frame.
Think of the folder you give Cowork as its working environment. What you put in that folder - before you start any task - determines how much Cowork already knows without you explaining it.
This is how you would implement this graph.
Create two folders before your next Cowork session:
Folder 1: “My Brief/”
This is a permanent folder that lives somewhere stable on your computer. It contains three files Claude reads at the start of every session to understand who you are:
- “about-me.md” - your role, what you do, what good work looks like for you
- “voice.md” - how you write and communicate (the same file from your Chat Project works here)
- “working-style.md” how you want Claude to behave. Do you want questions before it starts? Short or long outputs? Specific file formats?
# Working Style
## How I want Claude to work
- Always ask clarifying questions before starting any multi-step task
- Save outputs as .docx unless I specify otherwise
- Short section headers, not long explanatory titles
- Show me a plan before you execute it
## What I want to avoid
- Long preambles before the actual content
- Bullet-pointed everything — use prose where prose fits
- Generic advice — if you don’t know something specific, say so
## When I want to review
- Before deleting or overwriting any file
- Before sending anything external
- When you’re about to make more than 3 changes at onceFolder 2: Your actual project folder
Whatever you’re working on - a client project, a content series, a research project. This is where Cowork does its work. Files go in, outputs come out.
Point Cowork at your project folder when you want it to work. Point it at My Brief as a starting reference when you want to calibrate a new session. Or combine them - most of my sessions start with: “Read everything in the My Brief folder first, then let’s work on [specific task].”
Layer 2: Global Instructions - write once, applies everywhere
Global Instructions are standing orders that load at the start of every Cowork session, regardless of which folder you’ve selected. Most people skip this. It’s the highest-leverage setup step after the context files.
Go to Settings → Cowork in Claude Desktop. Click Edit next to Global Instructions.
What to write:
My name is [Name]. I’m a [role] who [what you do].
How I communicate: [2-3 sentences. Direct? Detailed? Do you prefer short answers or thorough ones?]
Output defaults: [File formats. Length preferences. How you want things structured.]
Always: [Things Claude should always do. Ask questions first. Flag uncertainty. Show plans before executing.]
Never: [Things to avoid. Specific phrases or patterns that sound wrong to you.]The difference between global instructions written in one paragraph and global instructions written in specific sections like the above is significant. Specificity is the whole game here. “Be professional” tells Claude nothing. “Never open a section with a gerund phrase” tells Claude exactly what to do.
Layer 3: Folder Instructions - per-project memory
Global instructions apply everywhere. Folder instructions apply to one specific folder.
This is useful for client work. Each client folder can have a brief that Cowork loads when you select it. Every session with that client starts fully briefed - no re-explaining who they are, what the project is, what tone they prefer.
To create folder instructions: open a Cowork session with that folder selected, then go to Cowork → Folder Instructions (right side of the chat) and add instructions for the current folder in the form of “CLAUDE.md”.
What to include:
Project: [Name of the project or client]
Context: [What this project is, where it stands, the goal]
Tone/style: [How outputs for this project should sound]
Constraints: [Anything to avoid, limits to stay within, formats required]
Files to know: [Key files in this folder and what they contain]One example from how I use this: I have a folder for newsletter production. The folder instructions tell Cowork my editorial calendar, my audience profile, the post format I use, and which files contain my voice rules. Every session I open on newsletter work starts with all of that already loaded. I just tell it what I’m working on today.
Layer 4: Plugins - pick one, ignore the rest
Plugins are specialist packs that add slash commands and domain knowledge for specific functions. Anthropic shipped eleven of them. Don’t install everything. Pick one that fits your role, install it, and learn what it adds before adding another.
To install: click the “+” button in the Cowork chat bar, then click Plugins.
Which one to start with:
- Productivity - good default if you’re unsure. Task management, daily planning, workflow structuring.
- Marketing - if you create content, run campaigns, or manage brand communications (screenshot below).
- Data Analysis - if you work with CSVs, SQL, exports, or any structured data.
I will cover Plugins and which ones to install later.
After installing, type “/” in a Cowork session to see the commands it adds. Run one. The output with a plugin active is noticeably more structured and specific than a generic prompt for the same task — it knows what a finished deliverable looks like for your function.
Layer 5: Connectors - the most underused feature
Connectors give Cowork live access to external tools. Google Drive, Notion, Slack, and fifty others. Once connected, Claude can pull data from them without you copying and pasting anything. I got mine connected to Notion because I open it every day.
Go to Settings → Connectors in Claude Desktop. Pick the one tool you use most. Connect it. Authenticate. Done.
You only do this once. After that, Claude has access to live data from that tool in every session.
The first prompt to run after connecting Google Drive:
Find the most recent document about [project name] in my Drive.
Read it and tell me the three most important things I need to know before we start working on it today.After connecting Notion (I personally use this one):
Search my Notion workspace for anything tagged as a blocker or flagged as urgent.
Summarize what needs my attention today.That’s it. No download. No copy-paste. Just an answer.
Your first action: Write global instructions. Create your My Brief folder with the three files. Point Cowork at a project folder. Open a new session and type:
Read the files in my My Brief folder.
Then tell me what you know about me, how I like to work, and what preferences you’re aware of.
Flag anything that’s missing or unclear.That’s your calibration test. If the response reflects how you actually work, you’re set up right. If something’s off, fix the relevant file.
Alright, now let’s move on to Claude Code.
