Back to Free Resources

Free Guide

How to Set Up Claude Cowork the Right Way

Most people open Cowork, type a few prompts, and wonder why it does not feel any different from regular Claude. The problem is the setup. This guide fixes that permanently.

What Cowork Actually Is

Regular Claude chat is a question-and-answer loop. You send a prompt, it responds, context resets next session. Cowork is different in a meaningful way: it has direct read and write access to folders on your computer, it plans and executes multi-step work autonomously, and it runs tasks while you do other things.

The simplest framing: chat is a tool you query. Cowork is something you delegate to.

That distinction changes what setup looks like. You would not hand work off to someone new without giving them context about your business, your preferences, and how you want things done. Cowork is the same. The five steps below are how you give Claude that context, and how you make it stick across every session.

One thing to know: Cowork launched as a research preview in January 2026, first for Max plan users and then for Pro, Team, and Enterprise shortly after. Features are still evolving, but the core setup described here is stable and working well across a wide range of workflows.

The Setup

Five Steps. In Order.

Steps two and three are where most people bail, and why most setups fall flat. Do not skip them.

01

Connect your tools first

Before you write a single instruction, connect your tools. Google Workspace, Slack, Notion, your project management system — whatever you actually use daily. The order matters: when Claude can access your existing documents during setup, it can draft your context files from what already exists. You type less and get better results. After connecting Google Drive, ask Claude to scan your about pages, bios, and company docs. It does most of the writing for you.

02

Create your three context files

Three markdown files. This is the step that makes everything else worth doing. about-me.md — who you are, what you do, who you serve, what success looks like in your role. brand-voice.md — how you write, what phrases you use, what you avoid, a few examples of your actual writing. working-style.md — ask first or just go, preferred file formats, how detailed responses should be, your hard rules. Store all three in a dedicated folder. Claude loads them at session start.

03

Set your global instructions

Condense your three files into roughly 800 words. Claude Desktop, Settings, Cowork, Edit. Paste and save. This layer loads before every session automatically — no folder selection required. The format: who you are, how you work, output defaults, voice and tone, business context, rules. Once this is in place, every session starts with Claude already briefed on who you are and how you want things done.

04

Install three to five skills

Skills are pre-built capabilities that give Claude something specific to do. Without them, Claude can chat. With the right skills installed, it can research and profile prospects, draft content in your voice, generate pre-meeting briefs from your calendar, create reports from your data, or run multi-step workflows end to end. Search inside Cowork by keyword — content, sales, research, project management. Pick three to five that match your most repetitive work. Add more once those are working.

05

Set up one scheduled task

A scheduled task is a prompt that runs automatically on a cadence you define. Claude executes it and delivers the output without you triggering it. Start with one: a morning inbox summary, a weekly task digest, a pre-meeting brief pulled from your calendar. Ask yourself what you do every day that takes five minutes and produces the same output. That is your first scheduled task. Get it working well. Then build from there.

What Good Context Files Look Like

The context files do not need to be perfect. They need to answer the questions Claude would otherwise have to guess at. Here is what each file should actually contain.

about-me.md

Start with a one-sentence description of what you do. Add who your clients or audience are, what industry you work in, what tools you use daily, and what good work looks like in your role. Two hundred words is enough to start. The file improves as you refine it.

brand-voice.md

This is the most underrated of the three. Without it, Claude writes in its default voice, which reads like AI. With it, Claude writes in yours. Include your tone (direct, conversational, formal), what words or phrases you use, what you hate hearing in your own copy, and a paragraph from an email or post you are proud of. That sample does more than a hundred rules.

working-style.md

How Claude should behave, not just what it should produce. Should it ask clarifying questions before starting, or attempt it and flag assumptions? What file formats do you prefer? How detailed should responses be? What are your hard rules? Example: "Never use em-dashes. Always show your reasoning. If uncertain, ask rather than guess."

Store all three in a dedicated folder. Select that folder at the start of sessions. That is how Claude loads them. The value compounds over time: each line you add or adjust makes the next session slightly better than the last.

Global Instructions: The Always-On Layer

Context files load when you select a folder. Global instructions load before every session regardless of which folder you have open. This is the condensed version of your context so Claude is never starting completely cold.

Take your three files and distill them into roughly 800 words. Structure it as: who you are, how you work, your output defaults, your voice, your key business context, and your rules. Claude Desktop, Settings, Cowork, Edit, paste and save.

Quick Verification

Claude Desktop
Tell me what you know about me and how I like to work.

If Claude can describe your role, communication style, and preferences without you saying anything, it is working.

Scheduled Tasks: Where It Gets Interesting

Once your context files and global instructions are in place, scheduled tasks become genuinely useful. Without that foundation, they just produce generic output on a timer.

Three starting points that work well: a morning inbox summary that flags urgent items, a weekly digest of completed work, and a pre-meeting brief generated from your calendar the night before. Write a specific prompt for each. Not "summarize my inbox" but exactly what you want flagged, how long the output should be, and where to save it. The more specific the prompt, the closer to done the output lands.

One constraint that matters: scheduled tasks require Claude Desktop to be open and your computer to be awake when the task runs. If you close the app, it does not run. Use scheduling for convenience automations where a missed run is not catastrophic.

Common Pitfalls

Five Mistakes Worth Avoiding

Skipping the context files

This is the main one. Without context files, every session starts from zero. You re-explain your role, your preferences, your style. Outputs are generic. The context files are not optional setup — they are what make the whole system worth using.

Connecting tools after writing instructions

Connect your tools first. When Claude can read your Drive during setup, it can pull from existing documents to help draft the context files. If you write instructions before connecting tools, you do all that work manually. Reverse the order and setup takes half as long.

Installing every available skill at once

Start with three to five. More options create more confusion and make it harder to debug when something goes wrong. Pick skills that match your most frequent work. Master those. Add more later.

Using Cowork for tasks that chat handles fine

Cowork uses significantly more compute per task than standard chat because of all the work running in the background. Reserve it for tasks that require file access, multi-step execution, or scheduled automation. For quick one-off questions, regular Claude chat is more efficient.

Building critical processes on scheduled tasks

Scheduled tasks require Claude Desktop to stay open and your computer to be awake. If either is not true, the task does not run. Use scheduling for convenience automations where a missed run is inconvenient but not catastrophic — not for anything time-sensitive or mission-critical.

What to Expect After Setup

The first session after a proper setup feels noticeably different. Claude does not need you to explain your role, your audience, or how you want things formatted. It already knows. Outputs land closer to done on the first attempt.

The value compounds as you refine the context files. Every time you add a line (a new rule, a preference, a detail about a workflow), the quality of outputs improves. What starts as a rough draft of your context becomes increasingly precise over time.

The setup takes about 30 minutes done manually. If you want to move faster, there are skills and plugins that interview you and build the context files, global instructions, and first scheduled task in under 10 minutes.

Zero to Automated

Get the full AI operating system

Guides are a great start. But inside Zero to Automated, you get a done-for-you AI system deployed within 48 hours, plus everything you need to keep building:

  • Claude Code 101 — 8 lessons, zero to automating real work by Sunday
  • Ready-to-install Claude Skills — a growing library with 2+ new skills every week
  • Done-for-you workflows — copy, paste, run. New ones every week.
  • Weekly live builds — watch real projects get built, ask questions, follow along
  • Small group coaching — direct access to both founders
See what's inside

$50/mo. Price locked for life at whatever you join at.