Someone on your team has probably tried Claude Cowork already, or pitched it to you as “the AI that does the work for you”. It does go well beyond a chatbot: it works with your files, your email and your tools. That’s exactly why, before you hand it to the whole office, you need to know what it is, what it costs and what it can touch.
The short answer
- What it is: Claude’s agent mode, from Anthropic. Instead of answering questions, it finishes multi-step tasks: reads a folder, cross-checks data, builds the spreadsheet or report and saves it.
- Pricing: it isn’t free. It’s included in Claude’s paid plans: Pro (from $17/month), Max ($100 or $200/month), Team (from $20 per user per month) and Enterprise. Prices are listed in US dollars, excluding tax.
- Where it runs: the Claude desktop app for Mac and Windows, plus web and mobile (still in beta on Team and Enterprise). To work with files on your computer, the desktop app has to be open.
- For a business: use Team or Enterprise, not personal accounts. You get admin controls (per-department permissions on Enterprise), and your data isn’t used to train models by default.
- The real risk isn’t the tool. It’s giving it access to files, email and systems with no usage policy and no training. Get those right and a large share of repetitive admin work can be delegated.
What is Claude Cowork?
Claude Cowork is the way of working with Claude where you delegate an outcome, not a question. You describe the task (“reconcile these invoices against the bank statement and flag the ones that don’t match”), Claude proposes a plan, carries it out step by step and hands you the finished file. You can watch each step, redirect it halfway or stop it.
Anthropic launched it in January 2026 as a research preview on Mac, made it generally available on macOS and Windows in April 2026 and brought it to web and mobile in July. It’s built on the same agent technology as Claude Code, but designed for office work rather than programming.
A recent change worth knowing about: on 16 September 2026 Anthropic announced that Cowork and chat are merging into one experience. On Pro and Max you no longer pick a “Cowork mode”: you type what you need and Claude decides whether a quick answer will do or a full task with files and tools is needed. Team and Free will follow; on Enterprise it’s a beta that stays off until an admin turns it on. That’s why Anthropic’s page now says “Claude Cowork is now just Claude”. The capabilities are the same; only the front door changes.
What’s the difference between Claude, Claude Cowork and Claude Code?
| Claude chat | Claude Cowork | Claude Code | |
|---|---|---|---|
| Best for | Questions, drafting, summarising | Multi-step office tasks | Writing and maintaining software |
| What you get | An answer on screen | Finished files: spreadsheets with working formulas, decks, documents | Code and repository changes |
| Works with | Whatever you paste or attach | Folders you choose, connectors, a browser, scheduled tasks | Code and the terminal |
| Who uses it | Anyone | Admin, sales, finance, HR, management | Developers and IT |
In a company they work together: the office runs on Cowork and IT speeds up its projects with Claude Code. That’s exactly the set-up in the case study below.
How much does Claude Cowork cost? Is it free?
No, it isn’t free. Anthropic says Cowork is available on paid plans only. It has announced that the unified experience will come to the Free plan too, but today Free doesn’t include Cowork. These are the published prices as of October 2026:
| Plan | Price | Who it’s for | Notes |
|---|---|---|---|
| Free | $0 | Trying the chat | No Cowork |
| Pro | $17/month (annual) or $20/month | Individuals | Cowork uses up your limit faster than chat |
| Max | $100/month (5x) or $200/month (20x) | Heavy users | More capacity for long tasks |
| Team Standard | $20/user/month (annual) or $25 (monthly) | Companies of 2–150 people | Includes Claude Code and Cowork, central admin |
| Team Premium | $100/user/month (annual) or $125 (monthly) | Daily, heavy users | 5x the usage of a Standard seat |
| Enterprise | $20/user/month (annual, 20-seat minimum) + usage at API rates | Organisations with security and compliance needs | Role-based permissions, SCIM, audit logs, Compliance API |
Prices exclude tax and can change, so always check Claude’s pricing page before you decide.
Two things the pricing tables don’t tell you:
- Usage limits matter more than the price. A Cowork task uses far more than a question, because every step calls the model. A Standard seat may not be enough for someone delegating work all day. On Team you can mix Standard and Premium seats by role.
- The real cost is adoption. Paying $20 per person while half the team doesn’t use it (or misuses it) is the most expensive option of all. We explain below how we avoid that.
How to use Claude Cowork: what it can do
These are the building blocks of work in Cowork, as Anthropic describes them:
- Folders. You give it access to a specific folder on your computer and Claude can read and write only there. You can add folder instructions with the project’s context.
- Connectors. Links to tools like Microsoft 365, Google Drive, Slack and others, built on MCP, the open standard for connecting AI to applications. See our guide to what MCP is.
- Skills. Reusable instructions that turn a repetitive task into a command. For example,
/quoteknows which template to use, where the prices come from and how to name the file. - Plugins. Bundles of skills, connectors and sub-agents by function (finance, legal, sales…). On Team and Enterprise, the company can run its own private catalogue.
- Scheduled tasks. Recurring work that runs on its own, in the cloud, on a schedule: Monday’s weekly report, a summary of pending emails every morning.
- Built-in browser. For web portals with no connector. Claude tries connectors first, then the browser and, as a last resort, computer use (in beta on Pro and Max), asking permission for each app.
- Projects and memory. Workspaces with their own files and instructions, plus memory shared between chat and Cowork.
- Artifacts. Documents, tables, charts or small interactive apps that Claude builds in a separate window, which you can edit, reuse and share with your team: a tracking dashboard, a quote calculator, a report ready to send.
Mac and Windows. It works the same on both, through the Claude desktop app. On Windows you need the latest version of the app. Web and mobile let you start and check tasks, but to touch local files the desktop app must be open and connected.
Getting started: open Claude, describe the task in terms of the result you want (“save a spreadsheet in this folder with…”), review the plan it proposes and let it run. On Team, and on Enterprise unless an admin has turned on the unified beta, you currently choose “Cowork” in the message box; on Pro and Max, with the new experience, you no longer need to.
Business use cases by department
These are the kinds of processes we’ve actually put into production in companies, not a theoretical list:
| Department | Task you delegate | How it works in Cowork |
|---|---|---|
| Sales | Answering quote requests | A /quote skill reads the email, applies the price list and leaves a draft for review |
| Sales | Following up on proposals | A /follow-up skill finds proposals with no reply and drafts the emails |
| Finance | Invoice reconciliation | Cross-checks the bank statement against the invoices in a folder and flags gaps in a spreadsheet |
| HR | Payroll checks | Compares the payroll provider’s PDFs with internal data and lists discrepancies |
| Operations | Monthly reports for clients or councils | A scheduled task gathers the data and leaves the report ready for review |
| Management | Weekly report | A /weekly-report skill summarises KPIs and open issues |
| Everyone | Email triage | Sorts the inbox, summarises what’s urgent and drafts replies, always reviewed before sending |
The rule we apply: Claude prepares, a person signs off everything that leaves the company.
Security and data: what Cowork can access in your company
This is the part most guides skip, and the part that matters most to a business.
- It reaches what you give it. The folders you choose, the connectors you switch on and the sites it visits. Within those, it can read, create and edit. Anthropic requires explicit approval before it permanently deletes files.
- Permission modes. In manual mode, Claude asks before each action. In auto mode, it approves read-only actions and judges write actions itself. A third mode asks nothing at all; Anthropic advises against it unless you trust everything involved.
- Tasks run in the cloud, on Anthropic’s servers, in an isolated environment (on Team and Enterprise this is in beta and the admin controls it). That isolation limits where code runs, not what Claude reads or does through your connectors, as Anthropic’s own guide to using Cowork safely points out.
- Prompt injection: an email, PDF or website can hide instructions that try to make the agent do something you never asked for. The risk grows when the agent reads outside content and also has permission to act.
- Admin controls on Team and Enterprise: switch Cowork and cloud sessions on or off (on Enterprise, per team through groups and custom roles), decide whether auto mode is offered, require approval for write-capable connectors, turn off web search and stream activity to your monitoring tools (OpenTelemetry). Enterprise can also scan skills and plugins at install time.
- Where data is processed. With a Claude account, data is stored in the US and requests can be processed in the US, Europe, Asia or Australia, with no option to keep them in the EU. If your data needs to stay in Europe, you need a different architecture; see our guide to EU data residency for AI models.
The risks of rolling it out on your own
Cowork is powerful precisely because it acts. Without a planned rollout, that creates specific problems:
- An agent with access to files, email and systems and no usage policy. Nobody has decided which folders can be connected, which data must not leave, or who reviews what goes out. Every employee sets it up their own way.
- Personal accounts instead of a company plan. An employee pays for Pro on their own card and connects their work email. The company sees nothing, can’t revoke access when the person leaves and, depending on their settings, those conversations may be used to train models.
- Connectors with write access. A connector that can change the CRM, send emails or move files turns a mistake, or a malicious email, into a real action. And “the AI did it” isn’t a defence: Anthropic is clear that you remain responsible for what Claude does on your behalf.
- Company data and credentials leaving the network. Connecting the ERP by pasting a password, or opening a port to the internet so the agent can reach it, is the quickest way to create a security hole.
- Untrained staff. Either they misuse it (feeding it data they shouldn’t, signing off without checking) or they don’t use it at all and you pay for licences that do nothing. And Article 4 of the EU AI Act, in force since February 2025 and softened by the Digital Omnibus in July 2026, still requires companies that use AI to take measures to build their staff’s AI literacy.
- A tool that changes every month. In nine months Cowork has gone from a Mac-only preview to merging with chat and running tasks in the cloud. What you set up in spring may behave differently in autumn if nobody keeps track.
A partner who has already rolled it out in a real company knows where these holes are before you fall into them: which architecture to choose, which permissions to grant, how to write the usage policy and how to train each department on its own work.
Want to roll out Claude Cowork without these risks? Book a free call and we’ll walk you through how we did it in a real company.
How we do it at Aitomatik
Our most complete Cowork rollout is with a multi-company services group in the Balearic Islands, where we act as their outsourced AI team. Here’s how it went:
- Discovery, one department at a time. We visited 8 departments and mapped 75 processes where AI or automation could help.
- Re-checking the plan as the tools moved. Looking closely at the 75 processes, we found that most of them could be done with Claude Cowork, with no custom development. The route to value became much simpler and cheaper.
- Making it compliant. Management had one non-negotiable: data and credentials could not leave the company. We walked them through every possible architecture and, together with their IT team, built a read-only gateway inside their own network: no ports opened, everyone sees only what they already could (Active Directory permissions), credentials never leave the gateway, and everything is logged. An employee asks “which invoices from this client are overdue?” and only the answer travels to the model.
- Training the whole office, in two rounds. First a general workshop, From chatbot to agent: delegating results rather than asking questions, turning repetitive work into skills (
/quote,/follow-up,/weekly-report), projects per area with their own files and instructions, connectors to their tools, reusable artifacts such as dashboards or calculators, scheduled tasks, email triage, and a clear usage policy (corporate accounts only, read-only access to systems, a person signs off everything that leaves the company). Then one session per department on its real processes: payroll checks with HR, quotes with sales, reports with the technical office, reconciliation with finance. Every team left with its own skills working. - Staying alongside them. Follow-ups with each department head after a few weeks of use, new automations, and backing for their IT team, which also uses Claude Code to move faster.
The result: more than 1,600 hours a year freed from repetitive admin work across the group, with data and credentials staying inside their network.
If you want the same, our AI training for companies and AI and automation consulting cover it. And if Cowork needs to reach your internal systems, we build the secure connector (an MCP server) for you.
Claude Cowork vs ChatGPT and Microsoft Copilot
- ChatGPT: OpenAI’s equivalent is ChatGPT Work, its agent mode for multi-step tasks. In the desktop app it can also use local files and desktop apps if your admin allows it, and ChatGPT Business has skills and custom MCP connectors too. If your company already runs on ChatGPT, see our comparison of ChatGPT vs Claude vs Gemini for business and our guide to ChatGPT Business.
- Microsoft Copilot: Microsoft has launched Copilot Cowork inside Microsoft 365 Copilot, built with Anthropic’s technology. It makes sense if your whole company lives in Microsoft 365 and already pays for Microsoft 365 Copilot; Cowork usage is billed on top, in Copilot Credits. Claude Cowork also has its own Microsoft 365 connector.
- Claude Cowork is strong at file work (spreadsheets with formulas that work, decks, documents), with skills, plugins and MCP connectors to your own systems. None of these is unique to Claude any more, so compare them on your real tasks.
The choice depends more on where your data and tools already live than on the brand. What matters is picking one and rolling it out properly.
Frequently asked questions
Is Claude Cowork free? No. It’s included in Claude’s paid plans (Pro, Max, Team and Enterprise). The Free plan doesn’t include it today.
Does Claude Cowork work on Windows? Yes. It has been generally available on macOS and Windows through the Claude desktop app since April 2026. On Windows you need the latest version of the app.
What’s the difference between Claude Code and Claude Cowork? Claude Code is built for programming and works with code and the terminal. Cowork uses the same agent technology for office work: documents, spreadsheets, email and reports.
Does Claude Cowork still exist? Yes, though its shape is changing. Since September 2026 Anthropic has been merging it with chat into a single experience, starting with Pro and Max. The capabilities stay: you ask for a task and Claude decides whether it needs to work with files and tools.
Is Claude Cowork safe for company data? It can be, with a Team or Enterprise plan, permissions limited to what’s needed, read-only connectors wherever possible, human sign-off for anything that leaves the company, and trained staff. Without those, the risk isn’t the tool but how it’s used.
Which plan does my company need? For a team, Team: central admin, usage analytics, OpenTelemetry monitoring, and data that isn’t used for training by default. Enterprise if you need per-department role-based permissions, SCIM, audit logs or the Compliance API.
Ready to roll out Claude Cowork in your company?
We’ve done it in a real company, from the first site visit to training the last department. We’ll help you choose the plan, design the architecture so your data stays where it should, write the usage policy and train your team on its own work. Book a free call and we’ll tell you where to start.
Claude Cowork pricing and features checked against Anthropic’s pages in October 2026. The tool changes often. Last reviewed: 15 October 2026.
