Claude Cowork for business: plans and set-up

Short answer: Claude Cowork is a mode of Claude that takes a goal, works through the steps on its own, and hands back finished files. It is included on every paid plan (Pro, Max, Ultra, Team and Enterprise) and not on Free. For a business the decisions that matter are which folders and apps it can touch, which permission mode people use, and what your admins switch on.

Below is what Cowork is, how it differs from chatting with Claude, what it can reach, where the risks sit and a sensible way to roll it out. Everything here comes from Anthropic’s own pages, checked on 31 October 2026.

What Cowork is

Anthropic describes Cowork as an agentic mode that uses the same architecture as Claude Code, aimed at knowledge work rather than software. You do not need a terminal. You describe an outcome, Claude plans the steps, may split the work into parts, runs them, and delivers documents, organised files or research you can preview and download.

Anthropic lists analysts, project managers, operators, researchers, marketers and recruiters as the people it suits, meaning anyone whose job involves moving information between systems. Typical jobs are weekly updates, meeting prep, catch-ups and triage.

How it differs from chatting with Claude

In a normal chat, Claude answers and you do the rest: copy the text into a document, build the deck, send the email. In Cowork, Claude does that work itself across your files and connected apps, and shows its steps as it goes. You can redirect or stop it at any time.

The line between the two is getting thinner. Anthropic says Chat and Cowork are merging into one interface where the change has rolled out, with every conversation starting in the same message box and Claude bringing in Cowork’s abilities when a task involves your files. Until it reaches your account, Cowork appears next to Chat.

Which plans include it

Cowork uses your usage allowance faster than chat, because every step of a multi-step task spends tokens. Group related work into one task and start a fresh task for unrelated work. For seat prices in Australian dollars, see our Claude pricing guide.

What it can do

• Work with files on your computer. In the desktop app, Claude reads and writes local files without you uploading or downloading anything. It can only see folders you connect.
• Use your connected apps. Connectors link Claude to tools such as Microsoft 365. Plugins bundle skills, connectors and sub-agents so a team can share a setup.
• Browse and fill in forms. Claude can open sites, read pages, click and type. The desktop app has a built-in browser separate from your own logins, or it can use Claude in Chrome.
• Run in the cloud. Cloud sessions run on Anthropic’s servers and keep going if you close your laptop. Tasks that need your local files need the desktop app to stay open.
• Run on a schedule. Scheduled tasks run in the cloud on a recurring basis, so the computer does not need to be awake.

Computer use, where Claude works through your screen, is a research preview. Anthropic says Claude prefers connectors first, then the browser, and uses the screen only as a last resort.

What to watch for

Anthropic is direct that Cowork carries risks chat does not, because it acts and can reach the internet. Its isolated environment protects your computer, but it does not limit what Claude can do with the access you grant.

• Permission mode. Manual asks you to approve each action. Auto approves read-only steps, lets Claude judge writes and deletes, and blocks actions its safety review flags. Skip checks nothing, and Anthropic says to use it only when you fully trust everything involved. On Pro, Max, Ultra and Team the default is Auto. On Enterprise it is Manual.
• Folder scope. Connect the one folder a task needs, not your whole drive. Claude needs your explicit approval before permanently deleting a file.
• Prompt injection and data leaving. A web page or document can carry hidden instructions. Anthropic says Auto mode checks for this and for data exfiltration, and tells you to review planned actions on sensitive files and to judge how far you trust any connector or website.
• Tasks that carry consequences. For anything involving money, messages sent as you or important files, stay close to the task or switch back to Manual.
• Retention. Sessions are saved to your account. Deleting a task removes it from your history at once and from backend storage within 30 days. Cloud sessions can draw on chat memory, and you can turn memory off before the first message of a task.

What admins can control

On Team and Enterprise, owners and admins have real levers. Anthropic’s pages list these:

• Turn Cowork off for the organisation, and manage access by team or role.
• Set spend limits and control tool permissions by department.
• Decide whether Auto mode is available at all, and whether new tasks start in it.
• Require approval for every write-capable connector action, which overrides “Always allow”.
• Disable web search for Cowork and Chat, or turn off Claude in Chrome.
• On Enterprise, stream activity to your security tools through OpenTelemetry.

One limit worth knowing: network egress settings do not apply to web fetch, web search, MCP connections or Claude in Chrome, so those need their own decisions.

Setting it up for a team

1. Pick one team and a few repeat jobs. Weekly reporting, meeting prep and inbox triage suit Cowork because the steps repeat and the output is easy to check.
2. Start in Manual. Move to Auto once people can see what Claude does with their folders and connectors. Leave Skip off.
3. Give each job its own folder. Put the inputs there, keep client records and finance files elsewhere, and add folder instructions so Claude knows the conventions.
4. Approve connectors one at a time. Set write actions to “Needs approval” until you trust the result. Our write-up on connecting HubSpot to Claude with MCP shows the limits we set.
5. Write down the rules. Say what data may go in, who reviews the output and what never runs unattended. Share repeatable steps as plugins or Claude Skills.
6. Check usage after two weeks. Settings > Usage shows who is hitting limits, which tells you whether a premium seat is justified.

For a bigger team, our guide to rolling Claude out to a 100-person business covers the people side.

Where we fit

We’re Certified in the Claude Partner Network. Our Claude for business page explains what we do, and our AI consulting work covers plan choice, permissions, policy and training before a team starts using Cowork on real data.

Frequently asked questions

What is Claude Cowork? Cowork is an agentic mode of Claude that plans and carries out multi-step work on your files and connected apps, then hands back finished output. It uses the same architecture as Claude Code but targets knowledge work and needs no terminal.

Which Claude plans include Cowork? Pro, Max, Ultra, Team and Enterprise include it. The Free plan does not.

Is Cowork different from Claude chat? In chat you get an answer and finish the job yourself. In Cowork, Claude does the job across your files and tools, and Anthropic is merging the two into one interface as it rolls out.

Is Claude Cowork safe for business data? It runs in an isolated environment and can only reach folders and tools you connect, but Anthropic warns that isolation does not limit what Claude can do with the access you give it. Start in Manual mode, connect narrow folders and review actions on sensitive files.

Can admins turn Cowork off? Yes. On Team and Enterprise, admins can disable Cowork for the organisation, control access by team or role, and decide whether Auto mode is allowed.

Sources

• Cowork help article, Claude Help Center, checked 31 October 2026, for plans, permission modes, file access, admin controls, retention and limitations.
• Claude Cowork, Anthropic, checked 31 October 2026, for platforms, admin controls and the unified Chat and Cowork rollout.
• What is Claude Cowork, Anthropic product guide, checked 31 October 2026, for how Cowork differs from chat.

---