6 min read
Claude Cowork is the closest thing a one-person business can buy today to a capable assistant who works inside your files, your browser and your apps, and it comes included with a paid Claude plan you may already have. It is also the AI tool most likely to do something you did not intend if you set it up carelessly. Both things are true, and this review covers both.
What Claude Cowork actually is
Cowork is Anthropic’s agent mode for everyday knowledge work. In Anthropic’s own words, it brings the approach of Claude Code (its programming assistant) to non-technical work “without requiring a terminal.” Instead of chatting back and forth, you describe an outcome, such as “sort last quarter’s receipts into folders by month and build me a spreadsheet of the totals,” and Claude plans the steps and carries them out.
The agent part is the key word. An agent is an AI that takes actions, not just one that writes answers. Cowork can open and edit files in folders you connect, use a built-in browser to read pages and fill in forms, call your other apps through connectors, and run tasks on a schedule. Anthropic took Cowork out of preview in April, and coverage at the time confirmed it was available across paid plans on both macOS and Windows.
What it can do for a solo business
According to Anthropic’s getting-started guide, the main capabilities are:
- Local file access. On the desktop app, Claude reads and writes files in folders you choose, with no uploading or downloading.
- Connectors. Links to other apps (your calendar, email, accounting or CRM tools) so Claude can look things up or take actions there. These connectors use MCP, the standard we explained in our beginner’s guide to MCP.
- Browser actions. Claude can open websites, read pages, click, type and fill in forms.
- Scheduled tasks. Recurring jobs that run in the cloud, so your computer does not need to be awake.
- Real office files. Spreadsheets and slide decks you can keep editing in Excel and PowerPoint.
- Projects, instructions and memory. Group related work, set standing instructions for a folder or for everything, and share memory with regular Claude chats.
In practice, the jobs where solo owners report the biggest payoff are the tedious, multi-step ones: tidying a messy downloads folder, turning a pile of PDFs into a summary table, drafting a weekly report from several sources, researching suppliers and building a comparison sheet, or preparing a monthly invoice list from your records. These are tasks you would hand to a virtual assistant if you had one.
Who it is for, and who should skip it
Good fit: consultants, freelancers, agency-of-one owners and small online sellers who spend hours each week moving information between files, websites and apps. If you already pay for Claude and your work lives on a Mac or Windows computer, you can try Cowork today at no extra cost.
Poor fit: anyone who mainly needs quick writing help (a regular chat does that with less usage), and anyone who is not willing to review what the agent plans to do. Cowork is powerful precisely because it acts. If you will click “approve” without reading, you are better off with a plain chat assistant.
What it costs
Cowork is not sold separately. It is included in paid Claude plans (Pro, Max, Team and Enterprise, plus the newer higher-usage tier), and it is not available on the free plan. Pro is $20 a month, or a little less billed yearly; check Claude’s pricing page for current prices on the other tiers.
The real cost is usage. Multi-step agent tasks use far more of your allowance than quick questions, because every step consumes tokens (the small chunks of text AI models read and write, which is how usage is measured). Anthropic’s own advice is to group related work into one task, start a new conversation for unrelated work, and say when you only need a quick answer. If you hit limits on Pro after a week of heavy use, that is your signal to either trim tasks or step up a plan. You can check where you stand under Settings, then Usage.
The safety settings that matter most
This is the section to read twice. Cowork gives you three permission modes:
- Manual: Claude asks before it acts. Start here.
- Auto: Claude reviews each action and blocks ones it judges unsafe.
- Skip: no checks at all; Claude acts without asking.
Anthropic’s guidance is direct: Cowork runs in an isolated environment that protects your computer, but that “does not limit what Claude can access through the permissions you’ve granted.” In plain English, if you connect your email and allow it to send, it can send. Anthropic specifically asks you to review planned actions involving sensitive files, money, or messages sent on your behalf. Permanent file deletion always needs your explicit approval.
The biggest risk for an agent that browses and reads email is prompt injection: hidden instructions planted in a web page or message that try to hijack the AI. Read our beginner’s guide to prompt injection before you connect your inbox.
Our recommended starting setup:
- Use Manual mode for your first two weeks.
- Connect one folder, not your whole drive. Create a “Cowork” folder and copy files into it.
- Connect apps read-only first where possible, and add sending or payment abilities later, one at a time.
- Write down what the agent is allowed to do without asking. Our AI autonomy ladder gives you a simple framework.
Limits worth knowing
- File access, browser use and computer use need the Claude desktop app open and connected, plus an internet connection throughout the task.
- Sessions cannot be shared with someone else, although individual outputs can.
- Some features remain desktop-only, including plugins that rely on local connectors.
- It is still an AI. It will occasionally misread a file, misjudge a total or take a long route to a simple result. Check its numbers the way you would check a new assistant’s.
The verdict
Claude Cowork is worth trying for any solo business already paying for Claude, because the trial costs nothing extra and the time savings on multi-step admin are real. Treat it like a new hire: give it a small, well-defined job, watch it work, then widen its responsibilities as it earns trust. Start with a task you already know how to do, so you can tell whether the result is right.
If you have tried Cowork already, which task saved you the most time? And which one did you take back?
Related reading
- What Is MCP? A Beginner’s Guide to the Connector That Lets AI Use Your Other Apps
- The AI Autonomy Ladder: How a Solo Business Decides What AI Can Do Without Asking
- Notion Agents Review: What the October Rebuild Actually Gives a One-Person Business
- Custom GPTs vs Claude Projects vs Gemini Gems: Where a Solo Business Should Keep Its AI Instructions
