Claude Cowork Demo: What Each Task Type Looks Like Before You Subscribe

Opt-in SaaS trials convert at a median 14%, with tools that reach user activation in under ten minutes converting three to five times higher (Pulseahead, 2026 SaaS Trial Conversion Benchmark). Claude Cowork skips the trial model entirely. You need a paid Claude plan, starting at $20/month for Pro, before you can run a single task.

Anthropic launched Cowork on January 12, 2026 as a Max-only feature at $100/month. Demand was loud enough that four days later, on January 16, Pro subscribers got access at $20/month. Five months in, there is still no free tier and no trial period. If you want to know what Cowork does before committing, you are limited to demos and walkthroughs.

YouTube covers the demo side well. Search "Claude Cowork" and you get dozens of screen recordings showing organized folders, finished spreadsheets, and compiled research briefs. What video walkthroughs tend to skip is the evaluation detail: how the permission dialog works, what the planning step actually shows you, how long a typical task runs before delivering output, and what the failure modes look like when a task hits a file it cannot parse.

This guide fills that gap with a written, searchable walkthrough of the complete Cowork task loop. Each of the four primary use cases gets a concrete breakdown: file organization, document preparation, research synthesis, and data extraction. If you already have a paid Claude plan, you can follow along in real time. If you are still deciding, the sections below show you exactly what the subscription includes.

Claude Cowork is included on every paid Claude plan. Pro costs $20/month and gives you access to the full Cowork feature set: local file access, sub-agents, scheduled tasks, plugins, and MCP connectors. Max 5x at $100/month and Max 20x at $200/month provide five and twenty times the usage capacity, respectively. The feature set is identical across tiers. Max buys you more task runs and longer sessions, not additional capabilities.

Team plans run $25 per seat per month with centralized admin controls. Enterprise pricing is custom and adds OpenTelemetry event streaming for SIEM integration.

Download and launch

Cowork runs inside the Claude Desktop app, not in a browser. Download the app from anthropic.com for macOS or Windows. Log in with your paid Claude account.

At the top of the interface, you will see two tabs: Chat and Cowork. Click Cowork. The view changes from a conversational chat window to a task-oriented interface with a prompt field at the bottom, a "Work in a Folder" checkbox, and suggested task categories like "Organize files" and "Crunch data."

Grant folder access

Check the "Work in a Folder" box and select the directory you want Claude to work in. A permissions dialog appears the first time you select a folder. You choose between one-time access (revoked when the task finishes) and "Always Allow" (persists across sessions).

Start with one-time access and a test folder containing non-critical files. This gives you a low-risk way to see Cowork's behavior before granting broader permissions. Once you trust the workflow, switch to "Always Allow" for folders you use regularly.

Every Cowork task follows the same five-step pattern. Understanding this loop is the fastest way to evaluate whether Cowork fits your work.

1. Describe the outcome

Type what you want done in plain language. Cowork prompts work best when you describe the end state rather than prescribing steps. "Organize my downloads folder by file type and remove duplicates" gives Claude enough information to plan an approach. "First list all files, then create folders for each type, then move each file" constrains Claude unnecessarily and often produces worse results.

Good prompts name the input (which folder or files), the transformation (organize, extract, compile, analyze), and the output format (PDF, spreadsheet, folder structure).

2. Review Claude's plan

Before executing anything, Claude presents a step-by-step plan showing what files it will read, what changes it will make, and in what order. If the plan includes destructive actions like deleting files or overwriting existing documents, those steps are flagged explicitly.

This is your review checkpoint. Approve the plan to proceed, modify specific steps, or reject it and re-prompt with different instructions. You are never committed to a plan just because you submitted a prompt.

3. Watch sub-task execution

Claude breaks the plan into sub-tasks and runs them, often in parallel through sub-agent coordination. One sub-agent might scan files for duplicates while another builds a folder structure. Progress indicators in the sidebar show each workstream as it runs.

You can interrupt mid-execution. Adding "skip anything in the /archive folder" or "change the output to CSV instead" redirects Claude without restarting. The ability to steer mid-task is one of Cowork's practical advantages over batch scripts that run to completion or fail.

4. Code runs in an isolated sandbox

All script execution happens inside an isolated virtual machine on your computer, separate from your operating system. Claude runs Python scripts, installs packages like pandas or matplotlib, and uses tools like LibreOffice for file conversions. The sandbox handles the processing while your local file system provides inputs and receives outputs.

This matters for the demo evaluation: Claude can do real computation (statistical analysis, chart generation, format conversion) without you installing anything manually. The VM spins up per task and tears down when the task finishes.

5. Review and refine the deliverable

Finished output lands directly in the folder you granted access to. Claude summarizes what it produced: files created, files moved, documents generated. You review the result in your file system using your normal tools.

If the output needs adjustment, follow up in the same session. "Move the summary section above the charts" or "add a column for year-over-year growth" refines the deliverable without re-running the full task.

The full loop typically takes between two and fifteen minutes. File organization with a few hundred files finishes in under five minutes. Research synthesis across multiple web sources can run ten to fifteen minutes as Claude reads, compares, and compiles.

Anthropic's official documentation lists four primary use cases for Claude Cowork. Here is what each one looks like when you run it.

File organization

Point Cowork at a cluttered directory. Claude scans every file, analyzes content and metadata, builds a folder structure based on what it finds, and moves files into the appropriate directories. In published walkthroughs, a 186-file downloads folder was sorted into 11 categorized subfolders with 27 duplicates detected through file hash comparison rather than filename matching alone.

This is the lowest-risk task type and the best starting point for your first Cowork session. Results are immediately visible (open the folder and see the new structure), and one-time access is sufficient since a single organization pass does not need persistent permissions.

Document preparation

Provide source materials and describe the output you want. "Turn these five meeting transcripts into a summary document organized by decision, action item, and open question." Claude reads each file, extracts relevant information, and assembles a structured document.

Supported output formats include PDF, Word (DOCX), and Excel (XLSX) with working formulas. DOCX output is generally more reliable than XLSX for documents with complex layouts. Excel works well for clean columnar data but can produce unexpected results with merged cells and multi-region layouts. If your task requires a polished spreadsheet with specific formatting, review the output carefully before sharing it.

Research synthesis

Describe a research question and Claude gathers information from the web, cross-references sources, and compiles a brief. In one published example, analyzing transcripts from hundreds of podcast episodes to extract the ten most important themes completed in approximately 15 minutes.

Research prompts perform better when they are specific and bounded. "Compare the pricing models of Figma, Canva, and Adobe Creative Cloud for teams under 20 people" produces a more actionable output than "Research design tools." Name the competitors, define the comparison criteria, and specify the output format you want.

Data extraction

Point Claude at contracts, invoices, reports, or scanned documents and describe the fields you want extracted. Claude reads the documents and outputs structured data as CSV, Excel, or JSON.

The advantage over dedicated extraction tools is flexibility: you describe fields in natural language instead of configuring templates. The tradeoff is throughput. A dedicated extraction platform processes thousands of documents per hour. Cowork handles dozens per session. For high-volume extraction, a tool like Fastio's Metadata Views fills the gap. You describe fields in natural language, AI designs a typed schema, and extraction runs across entire workspaces of documents on persistent cloud storage, making extracted data immediately available to your team without file transfers.

Give Cowork a persistent workspace that outlasts your laptop

Fastio's Business Trial includes 50GB storage, five workspaces, and MCP server access. Connect it to Claude Cowork so task outputs persist in the cloud, accessible to your whole team. No credit card, no expiration.

Start 14-Day Trial

The walkthrough above covers single tasks. Cowork also supports recurring workflows through three features that extend the basic task loop.

Cowork Projects

A project bundles related tasks, files, instructions, and memory into one persistent workspace. Create a project for a recurring workflow, and Claude retains context across sessions. You do not re-upload the same brief or re-explain your preferences each time you start a new task within that project.

Projects support instruction files (CLAUDE.md) at the root and subfolder level. Drop a CLAUDE.md file at the project root with standing context like "always output reports as PDF" or "use AP style for all written content." Subfolder instructions extend or override the root-level ones. This replaces the repetitive preamble most users paste into each new chat session.

Scheduled tasks

Type /schedule in any Cowork session to create a recurring task. The scheduler accepts natural-language time descriptions: "every weekday at 8am," "first Monday of each month," or "every Friday at 5pm." Common setups include morning email summaries, weekly metrics digests, and regulatory update checks.

The constraint to know before subscribing: scheduled tasks only run while your computer is awake and the Claude Desktop app is open. If your laptop is closed at the scheduled time, the task does not execute and does not retry. For reliability-critical workflows, you need either a dedicated machine or a cloud storage layer that persists outputs regardless of machine state.

Fastio workspaces address this gap. Files written to a Fastio workspace by any agent persist whether the originating machine is on or off. Connect the Fastio MCP server to Cowork, and task outputs land in cloud storage that your team can access from any device. The Business Trial includes 50GB, included credits, and five workspaces with no credit card required.

Mobile Dispatch

Pro and Max subscribers can assign tasks from the Claude mobile app. Describe what you need on your phone, and Claude executes it on your desktop with full access to local files, plugins, and connectors. The same desktop-awake requirement applies: your machine must be on and the Claude Desktop app running for dispatch to work.

Dispatch is useful for time-sensitive requests when you are away from your desk. "Compile this morning's sales data into the weekly report template" sent from your phone at 7am can be finished before you sit down.

Cowork delivers outputs to the local folder you granted access to. For individual work, this is straightforward: open the folder, review the files, move on. For teams, it raises a sharing question that the demo walkthrough should address.

Local sync options

The simplest approach is syncing your Cowork output folder with a cloud storage service. Google Drive, Dropbox, and OneDrive all support folder-level sync, so files Claude writes locally appear in the cloud automatically. The limitation is that these services provide file storage without content intelligence. Your synced files are searchable by filename and basic metadata, not by meaning.

MCP-connected cloud workspaces

Cowork supports MCP (Model Context Protocol) connectors that link it to external services during task execution. When Claude writes to an MCP-connected workspace, outputs land in cloud storage without a local sync step.

Fastio's MCP server exposes workspace, storage, AI, and workflow tools that Claude calls during task execution. Connect it through the Cowork connector interface, and Claude reads source materials from and writes outputs to Fastio workspaces directly. Intelligence Mode auto-indexes uploaded files for semantic search and AI-powered chat, so research briefs and extracted data become queryable by your team the moment they land.

This matters for the demo evaluation because it shows Cowork is not limited to local-only workflows. An MCP connector turns Cowork into a front end for persistent, team-accessible cloud workspaces where agents and humans share the same files.

The Business Trial covers enough for a realistic evaluation: 50GB storage, five workspaces, and included credits for AI queries and document processing. Sign up at fast.io/storage-for-agents with no credit card and no trial expiration.