The step-by-step guide to setting up your AI-powered workspace. No technical experience required. From download to your first real task in under 30 minutes.
Most people know Claude as a chatbot: you type a question, you get an answer. It also does the work. Instead of answering one question at a time, Claude takes on actual work: writing documents, building spreadsheets, researching companies, making presentations, analyzing data, all as real files on your computer. Anthropic launched this as a separate mode called Cowork and, in September 2026, began folding it into the main app, so you will see both names for a while.
| Asking Claude a question | Handing Claude a task |
|---|---|
| You ask, Claude answers | You assign a task, Claude does the work |
| A reply in the conversation | Creates real files: Word, Excel, PowerPoint, PDF |
| Works from what you paste in | Reads and writes to a folder on your computer (desktop app) |
| Done in seconds | Can run for minutes while you do something else |
Memos, proposals, reports, emails, and posts as polished Word docs or PDFs.
Deep-dive a company, person, or topic and get a sourced research document.
Real PowerPoint decks you can open in PowerPoint or Google Slides.
Excel files with formulas and charts. Analyze a CSV. Build a dashboard.
Get the desktop app at claude.com/download, install it, and sign in with your Anthropic account. You type every request into the message box in the middle of the window. (Ignore the Code option, that's for software developers.)
Heads up: the first time you hand over a task that works with files, the app downloads a ~2GB environment. This happens once and can take 5-10 minutes. Let it finish.
Claude asks you to choose a folder. This is the one setup choice worth thirty seconds, because moving it later is a chore: create the folder inside your cloud-synced drive, not a plain local folder. Something like Google Drive/CoWork OS. Use whatever your work already runs on: Google Drive, Dropbox, OneDrive/SharePoint, or Box. Claude can only see files inside this folder, not your whole computer.
Type a request in plain English and be specific: "Write a one-page executive summary of my company for investors. Make it a Word doc." Claude asks a clarifying question or two if it needs to, works through the task while you watch or switch apps, then hands you a link to the finished file already sitting in your folder. No downloading.
You do not need any setup beyond Step 2 to get value. Paste one of these into Claude and watch it produce a real file. Start at the top, it needs nothing from you.
Message box shows a Chat / Cowork switch? Pick Cowork for this. Why.
"Research [Company Name]. Give me a Word doc covering what they do, their business model, key people, recent news, and competitive positioning."
Zero setup. Claude uses web search and hands you a real .docx.
"Write a one-page executive summary of my company for potential investors. Here's what we do: [brief description]. Make it a Word doc."
"Create a 10-slide pitch deck for [your company]. Include problem, solution, market size, business model, team, and ask. Make it a .pptx file."
Drop a CSV or Excel file into your folder, then ask: "Analyze the data in [filename]. Tell me the key trends and create a summary dashboard."
Once you're doing more than one-off tasks, use Projects, a built-in feature (since March 2026). A Project is a persistent workspace with its own files, instructions, and memory. Everything inside stays scoped to that Project, so Claude remembers your context across tasks without you re-explaining anything.
In Claude Desktop, click Projects in the left sidebar, then New Project.
Give it a clear name ("WuunderFund", "Q2 Marketing"). In the Instructions field, add a few lines on how Claude should work here: tone, formatting, key context. These apply to every task in the Project.
Add a local folder, link a chat project, or paste a URL for Claude to reference. This is what Claude draws from when it works.
Type your request. Claude now has your instructions and context loaded automatically.
Projects handle the basics. If you want a fuller system, an identity file that spans every project, a folder for each area of your work, rules Claude reads before every task, explicit memory you control, you have two options: build it by hand, or let Claude build it from one pasted prompt. Start with the prompt.
Open the Cowork OS setup prompt and copy all of it. He walks through it in this 16-minute video from June 2026, which is worth watching once.
Open the Claude desktop app with the cloud folder you chose in Step 2 selected. The prompt tells Claude to build inside "the folder you have access to," so the folder has to be connected first.
Message box shows a Chat / Cowork switch? Pick Cowork for this. Why.
Claude introduces itself and asks its questions in small batches. Before it writes anything, it shows you the folder list. Change what you want, then tell it to go ahead.
| Claude asks you… | What it builds |
|---|---|
| What should I call you? | A name and personality, written into the main CLAUDE.md |
| Your name and what you do? | About Me/Profile.md, so Claude knows who it is helping |
| Which areas do you want help with? | One subfolder per area, each with its own CLAUDE.md |
| How should I talk to you? | About Me/Voice.md |
| Anything I should never do? | About Me/Rules.md |
| What would make this useful? | The priorities in your main CLAUDE.md |
When it finishes, your cloud folder holds a main CLAUDE.md that points to everything else, a MEMORY.md index, an About Me folder with three files, and a subfolder for each area you picked. From then on, "remember this: [fact]" gets saved to memory, and "add an area for [name]" gets you a new subfolder with its own instructions.
/begin. If you set up that way, it still works and you do not need to redo anything. The March 2026 plugin file is still here. For plugins in general, see Claude Plugins.
A connector gives Claude direct access to a tool, no copy-pasting, no exporting. Adding one takes about 30 seconds: click "Connect" and sign in. Anthropic's directory lists more than 2,000 connectors and plugins. The ones most people start with:
| Connector | What Claude Can Do |
|---|---|
| Google Drive | Read and reference your Docs, Sheets, and Slides |
| Gmail | Read, draft, and send email; search your inbox |
| Google Calendar | Read your schedule, find conflicts, prep for meetings |
| Slack | Read channels, search messages, catch up on what you missed |
| HubSpot / Salesforce | Pull CRM contacts, deals, and account records |
Full directory at claude.com/connectors. All free. Add one under Customize → Connectors.
Beyond connectors, plugins teach Claude to do role-specific work (Sales, Marketing, Finance, Legal, Data, and more), each bundling skills and templates built by Anthropic. Browse them under Customize, see Claude Plugins for what to install and what to skip, and the Advanced Guide for building your own.
Claude comes in four models. You pick yours in the message box. From fastest to most capable:
Haiku 4.5, the fastest and cheapest. Reach for it on quick, simple jobs: reformatting text, a one-line question, high-volume back-and-forth.
Sonnet 5.5, the everyday pick. Fast and lower-cost, built for well-scoped everyday tasks and polished documents. It handles most of what you'll throw at it. Start here.
Opus 5.5, the deep-work model and the cost-effective pick for hard tasks. Best when the work is complex, open-ended, or runs long. Anthropic says it performs at the level of Fable 5.1 on most work, and it costs less. Make it your default when a job is genuinely hard.
Fable 5.1, the most capable model Anthropic ships. Built for the hardest reasoning and long-running problem-solving. It costs more per token than Opus, so save it for the roughly 10% of tasks where Opus gets stuck.
Rule of thumb: start with Sonnet 5.5. Move up to Opus 5.5 when the task is high-stakes, complex, or long-running. Drop to Haiku 4.5 for something quick. Escalate to Fable 5.1 only when Opus can't crack it.
A new task should be a new session. Long threads make Claude re-read everything and burn your limit (that's the warning from Step 1). When a job is done, start a clean one.
Role, task, format, audience. "A 10-slide investor deck for angel investors, confident tone, with a projections slide" beats "make a deck" every time.
The Drive connector reads Office files and Google Docs from a phone session, but not .md. If you want to pull something up on mobile, don't save it as markdown.
Claude reads your folder before every task. Don't dump files in "just in case", a tight folder means Claude reads fast and reads well.
| You want to… | Go to |
|---|---|
| Get dramatically better results from any prompt | The Prompting Guide |
| Build workflows Claude runs for you | Claude in Action |
| Set up the folder/CLAUDE.md/memory system by hand, build skills and plugins | The Advanced Guide |
| Hand off work from your phone and the web | Claude Everywhere |
| Decide which plugins to trust | Claude Plugins |