Guide

What is Claude Cowork? A plain-English guide for business leaders

Claude Cowork is Anthropic's agent for knowledge work: you describe an outcome, and Claude plans the steps, works across the folders and connected apps you allow, and hands back finished work such as documents, spreadsheets and slide decks for your review. It became generally available on all paid Claude plans on 9 April 2026 and has no price of its own. It is included in Pro, Max, Team and Enterprise, and it uses a plan's usage limits faster than chat does. On 16 September 2026 Anthropic began merging Cowork and chat into one Claude, starting with Pro and Max; Team and Enterprise organisations keep chat and Cowork as they are for now, and Enterprise admins get at least 30 days' notice of any change. That is how Anthropic's own pages described it when we checked on 29 September 2026. OpenAI's ChatGPT Work and Microsoft's Copilot Cowork do similar work, licensed and metered in different ways; our comparison sets them side by side.

Claude Cowork. An agent in Anthropic's Claude that carries out multi-step knowledge work, such as research, analysis and document creation, across the files and connected tools a user allows, and returns finished work for review.

Written by Tayyeb Mahmud, founder and Principal Architect. Claude Certified Architect – Professional · LinkedIn profile of Tayyeb Mahmud

1AYM is an OpenAI Select Partner.

Checked 29 September 2026. Every product fact on this page comes from Anthropic's, OpenAI's or Microsoft's own pages and help articles, read on this date. Prices are Anthropic's US dollar list prices. Anthropic says its US dollar prices do not include applicable tax. All three vendors change products, names, plans and prices without notice, so confirm them on the sources below before you buy.

What Claude Cowork does

Chat answers a question and stops there. Cowork takes a piece of work off your desk: you give Claude a goal and access to the folders and apps it needs, and it makes a plan, shows each step as it goes, and returns the result for you to review [1, 5]. Anthropic describes it as the approach behind Claude Code, its coding agent, brought to knowledge work beyond coding, with no terminal required [5].

If the work you have in mind is software, you want a different product. Cowork is for everything else; Anthropic sells Claude Code for engineering, and our comparison of Codex and Claude Code for business covers that choice [1].

Finished files, not text to paste: Formatted documents, Excel spreadsheets with working formulas and PowerPoint presentations, which you can open, edit and download [4, 5].
Work across your apps: Connected apps such as Google Drive, Gmail, Microsoft 365 and Slack, and, in the desktop app, a built-in browser in which it can read pages and fill in forms [1, 4].
Tasks that keep running: Tasks can run in Anthropic's cloud (in beta), so they carry on when the laptop is closed, and can be checked or redirected from a phone [1, 5].
Scheduled work: A task can be saved and run on a daily, weekly or monthly schedule, such as a Monday readout built from last week's figures [1, 5].
Anthropic's own examples: A recurring marketing report, reconciling regional spreadsheet exports against a budget, a first-pass review of inbound contracts against a legal playbook, and a memo on every open sales deal drawn from call notes [1].

Where Cowork stands in September 2026

Anthropic announced Cowork as generally available on all paid plans on 9 April 2026, and shipped with it controls for rolling it out across a company: role-based access on Enterprise, group spend limits, usage analytics and activity data for security tools [2].

On 16 September 2026 Anthropic began merging Cowork and chat into one Claude. In the new experience there is no mode to pick. You ask for what you need, and Claude decides whether that is a quick answer or a task, so everything Cowork did is available from any conversation [3, 4]. Existing Cowork tasks, projects, connectors, skills and files carry over [4].

The change reaches plans in stages. Pro and Max went first, on web, desktop and mobile. Anthropic says Team and Free plans will follow soon, and Enterprise admins will hear at least 30 days before anything changes for their organisation [3]. For now, Team and Enterprise organisations keep chat and Cowork as they are [6]. If you have read that Cowork is going away, that is the wrong conclusion to draw. The name is still on Anthropic's product and help pages, and the capability has been folded into Claude rather than withdrawn.

Which plans include Cowork, and what it costs

Cowork has no price of its own. It is included in paid Claude plans and is not on the Free plan [5, 7], so what you are really buying is the plan, and Cowork draws on that plan's usage. Expect it to use that allowance up faster than chat: Anthropic says agentic tasks consume more capacity, because Claude coordinates several sub-agents and tool calls to finish a piece of work [1]. If the plan itself is still open, our comparison of ChatGPT Enterprise and Claude Enterprise sets out price, controls and data location for each.

Claude plans that include Cowork, with Anthropic's published US dollar prices, which Anthropic says exclude applicable tax, as read on 29 September 2026
Dimension	Price	Who it is for
Pro [7]	US$17 a month billed annually (US$200 up front), or US$20 billed monthly	One person; Anthropic suggests it for quick tasks such as organising a folder or a short report
Max [1, 7]	From US$100 a month (Max 5x); US$200 a month for Max 20x	One person who hands work to Claude throughout the day
Team [7]	US$20 a standard seat a month billed annually, US$25 monthly; premium seats with five times the usage US$100 or US$125	Teams of 2 to 150
Enterprise [7]	US$20 a seat a month, billed annually, plus usage at Anthropic's API rates	Larger organisations that need SCIM, audit logs, a Compliance API and spend limits by user and organisation

UK buyers should confirm the price in pounds at checkout. Check the training default as well. On Team and Enterprise, Anthropic lists no training of its models on your content by default; on Free, Pro and Max its plan table lists model training as opt-out [7]. That difference alone is why I would keep company files off personal Pro and Max accounts.

What admins control on Team and Enterprise

Cowork is switched on by default for Team and Enterprise organisations, so unless someone changes it, the first decision has already been made for you. An owner can turn it off for everyone, and the settings below decide how far it reaches once it is on [6].

Who gets it: An organisation-wide on and off switch on both plans. On Enterprise, admins can put users into groups, by hand or through SCIM from the company's identity provider, and give each group a role that includes or excludes Cowork [2, 6].
Where tasks run: Running tasks in Anthropic's cloud has its own switch: on by default for Team, off by default for Enterprise until an owner enables it for chosen groups [6].
How much Claude may do unasked: Admins decide whether staff may use the mode in which Claude approves its own actions after a safety check. Letting staff skip approval for connector tools that can write data is off by default [5, 6].
What each connector may do: Admins can limit the actions available in each connector across the organisation, for example allowing reads but not writes [2].
Spend: Enterprise admins can set budgets per team from the admin console, and spend limits per user and for the organisation [2, 7].
Visibility: Usage analytics for admins, activity streamed to the company's security tools through OpenTelemetry on Team and Enterprise, and Cowork sessions captured in the Compliance API on Enterprise [2, 6, 7].
Plugins: Owners can publish an approved set of plugins, bundles of skills and connectors, and mark each one as installed by default, optional, required or hidden [6].

How it relates to ChatGPT Work and Copilot Cowork

OpenAI's ChatGPT Work and Microsoft's Copilot Cowork do the same kind of work, and Microsoft says it built Copilot Cowork working closely with Anthropic, integrating the technology behind Claude Cowork [15]. Where each one runs, how it is billed and what admins control are set out in our comparison of ChatGPT Work and Claude Cowork, which covers Copilot Cowork for firms on Microsoft 365.

Before you switch it on

An agent that can act across your files and apps needs a few decisions made before its first task, whichever vendor you choose. Each item below points to the vendor setting behind it, and the ones on access and spending take the most planning.

Pick one weekly task: Choose work someone does every week, such as a status report or a batch of documents to review, and time it now. Without a baseline, a pilot produces opinions rather than evidence.
Decide what it may reach: Grant the folders and connectors that task needs and nothing more. Anthropic advises against giving Cowork access to files with sensitive information and warns that the risk of prompt injection, instructions hidden in content it reads, is not zero [6, 16].
Keep approvals on for actions: Claude asks before acting by default [4], and Copilot Cowork asks before it sends an email or posts a message [10]. Keep it on for anything that sends, deletes or pays. It is the same principle as our pattern for gating an agent's consequential actions.
Set a spending limit before the pilot: Group spend limits on Claude Enterprise [2], per-user and per-group credit limits on Copilot Cowork [12], and workspace credits on ChatGPT plans [9]. On Copilot Cowork, remember that being inside a spending policy is what grants access, so even a one-credit limit lets a user in [13].
Decide where tasks run: Cloud tasks keep going when a laptop closes. Local Cowork sessions keep their history on the user's computer, where Anthropic's standard retention does not apply and admins cannot manage or delete it centrally [6].
Tell people what is changing: Pro and Max users are moving to one combined Claude. Anthropic says Team will follow, and Enterprise admins get at least 30 days' notice of any change [3]. A short note on what the tool may and may not be used for saves a lot of guesswork.

Where 1AYM fits

Setting up and rolling out tools like these is core 1AYM work: Claude Cowork, ChatGPT Work and Copilot Cowork. The closest documented case is engagement file D-01: at a large international marketing agency we hold the lead architect role on its AI platform, contribute engineering to its internal Claude Code skill platform and built its identity and provisioning pipelines.

1AYM is an OpenAI Select Partner, and I hold Anthropic's Claude certifications, each verifiable on Credly.

If you are still deciding whether to buy one of these agents or build your own, our AI Opportunity & Feasibility Sprint settles that over a typical two to four weeks. Once the decision is made, a fixed-scope rollout can start within a day of the scope being signed. The next step is a short call, booked from the end of this page.

For IT and security reviewers: where tasks run and what is logged

What a security review will ask about, each point taken from the vendor's own documentation. Where a vendor's pages contradict each other, the point says so.

Claude: cloud and local sessions: Cloud sessions run on Anthropic's servers in an isolated environment, and sessions and files are saved to the member's Claude account. Local sessions run on the user's computer, with code in an isolated virtual machine [5, 6].
Claude: local history: Local sessions store conversation history on users' computers, outside Anthropic's standard retention. Admins cannot manage or delete it centrally. Enterprise admins can retrieve it through the Compliance API; deletion endpoints for local sessions are not available yet [6].
Claude: network egress: Cowork respects the organisation's network egress settings, but they do not apply to web fetch, web search or MCP connectors, including Claude in Chrome. Owners can turn web search off separately [5, 6].
Claude: approval modes: Manual asks before each action. Auto approves actions after an automated safety check and consumes more usage. Skip runs without asking or checking. Admins can remove Auto, and members cannot set write-capable connector tools to "Always allow" unless an admin permits it [5, 6].
Claude: OpenTelemetry events: Events cover tool and connector calls, files read or modified, skills used, and whether each AI-initiated action was approved manually or automatically. Anthropic says this does not replace audit logging for compliance [2, 6].
ChatGPT Work: access controls: Work Cloud and Work Local are separate role permissions; Work Local without Work Cloud allows desktop work but no cloud tasks. Codex Local is controlled separately, and browser use and network access keep their own controls. Enterprise admins must enable event-triggered tasks before members can create them [8].
Copilot Cowork: access is a spending policy: At general availability, access comes only from a spending policy that selects Cowork. A more restrictive policy does not override a more permissive one, and credit limits are enforced asynchronously, so a user may start tasks after reaching a limit [12, 13].
Copilot Cowork: data boundary and regions: Microsoft says Cowork processes a user's files in a temporary, isolated environment inside the Microsoft 365 service boundary and follows Copilot's data residency model. Its use of Anthropic models is limited to the regions Anthropic supports [11, 12].
Copilot Cowork: Anthropic model setting: Microsoft's pages disagree. The model page says the Anthropic model setting is tenant-wide; the admin page says it can be scoped to users or groups [12, 14]. Confirm with Microsoft before a review relies on either.

Sources

Common questions

Is Claude Cowork free?

No. Anthropic lists Cowork on its paid plans only: Pro, Max, Team and Enterprise. Checked on Anthropic's pages on 29 September 2026.

How much does Claude Cowork cost?

Cowork has no separate price; it comes with the Claude plan. On Anthropic's pricing page, Pro is US$17 a month billed annually or US$20 monthly, Max starts at US$100 a month, Team is US$20 a standard seat a month billed annually or US$25 monthly, and Enterprise is US$20 a seat a month billed annually plus usage at API rates. Anthropic says these dollar prices do not include applicable tax, and Cowork uses a plan's limits faster than chat. Checked on 29 September 2026.

Is Claude Cowork being discontinued?

No. From 16 September 2026 Anthropic began merging Cowork and chat into one Claude, so what Cowork does becomes available from any conversation rather than a separate mode, and existing tasks, projects, connectors and skills carry over. It reaches Pro and Max first. Team and Enterprise organisations keep chat and Cowork as they are for now, and Anthropic says Enterprise admins will hear at least 30 days before anything changes.

Can Claude Cowork use files on my computer?

Yes, through the Claude desktop app for macOS or Windows, and only in the folders you connect. Claude asks before permanently deleting any file. On the web and mobile it works with files saved to your Claude account and your connected apps, and reaches your computer only while the desktop app is open.

Is Claude Cowork safe for business?

It can be, if access is scoped to the task and approvals stay on for actions. Anthropic's safety guidance, read on 30 September 2026, names prompt injection as the main risk: instructions hidden in a file, email or web page that Claude reads during a task. An attack needs two things at once, content from outside your trusted boundary and the ability to take actions that could harm you, so limit both. Anthropic advises against giving Cowork local files with sensitive information, such as financial documents, and says the chance of an attack is still not zero. Claude always asks before permanently deleting files, in any approval mode. Keep approvals on for anything that sends, writes or pays, and limit connectors to reading unless a task needs more.

Can we turn Claude Cowork off for our organisation?

Yes. On Team and Enterprise it is on by default, and an owner or primary owner can turn it off for the whole organisation in Organization settings. Enterprise admins can also grant it only to chosen groups through custom roles.

Does Anthropic train its models on what Cowork sees?

On Team and Enterprise, Anthropic lists no model training on your content by default. On Free, Pro and Max its plan table lists model training as opt-out. Check the setting on your own account before connecting work files.