Claude Cowork for UK business: what it is, what it costs, and ten jobs to hand it first

Most UK firms met Claude as a chat window. You ask, it answers, you copy the answer somewhere useful. Claude Cowork is the other half of the product, and for a business it is the half that matters: you describe the outcome you want, Claude works through your files, your connected systems and the web, and comes back with the finished thing. This guide covers what Cowork is now that Anthropic is folding it into every conversation, which plan a UK firm needs and what that costs, ten jobs worth handing it first, and the settings an owner should change before anyone starts. Every vendor fact was checked against Anthropic's own pages on 24 October 2026; the sources are listed at the end.

What Claude Cowork is

Anthropic's own description is the clearest: Cowork uses the same agentic architecture that powers Claude Code, with no terminal required. Instead of responding to one prompt at a time, Claude takes a multi-step task and executes it. In practice that means it plans the steps, breaks the work into sub-tasks and runs several in parallel, reads the folders and connected apps you have allowed, runs code and shell commands in an isolated environment on Anthropic's servers, opens websites in a built-in browser (or your own Chrome through the Claude in Chrome extension) when a task needs the web, and delivers finished outputs you can preview and download: Word documents, Excel files with working formulas and multiple tabs, PowerPoint decks, formatted reports.

The timeline matters because the product has moved quickly. Cowork became generally available on the desktop app on 9 April 2026. From 7 July it reached web and mobile, with sessions running in the cloud so a task started at a desk can be checked from a phone. On 16 September Anthropic began rolling out what it calls one Claude: chat and Cowork merged into a single conversation with no mode to choose. A quick question still gets a quick answer; a bigger brief becomes a task that keeps running after you close the laptop. That rollout started with Pro and Max plans. On Team it is in beta and appears only where an owner turns on the setting Anthropic names Chat and Cowork unified; on Enterprise it is available where enabled. If your message box still shows separate Chat and Cowork options, you have the earlier experience, and everything in this guide still applies.

Two architectural facts shape everything else. First, sessions run in the cloud by default. Claude's work happens in an isolated, temporary environment on Anthropic's servers that is created for the session, cannot reach your network and is removed when the session ends. When a task needs a local file or your browser, Claude reaches your computer through the open desktop app, and only for the folders you have connected. Anthropic is explicit that this means the work, including any local files opened that way, is processed on its servers rather than staying on your machine. Second, scheduled tasks run in the cloud with no device online at all, which is what makes the Monday-morning report possible and is also why Anthropic's safety guidance treats scheduled tasks with extra care.

What it can do that chat cannot

The capabilities Anthropic lists in its Help Center fall into five groups, and it is worth knowing them by name because each one changes what you can hand over.

• Files and folders. On desktop, Claude reads from and writes to local folders you connect, without uploads or downloads. Organise a Downloads folder by type and date, turn a folder of receipts into an expense report, batch-rename files to a date convention.
• Finished deliverables. Spreadsheets with working VLOOKUPs and conditional formatting rather than CSVs that need fixing, slide decks from rough notes, reports from voice memos and scattered notes. Anthropic's Claude Docs, Slides and Design tools, in beta, produce documents and decks you can edit in place and export to Word, Google Docs or PowerPoint.
• Connected apps. Connectors to Google Drive, Gmail, Microsoft 365, Slack, HubSpot, Xero and others work inside a task, so Claude can pull what it needs as it goes rather than waiting for you to paste it. Our earlier guide on connecting Claude to Xero and HubSpot covers the scopes each one asks for.
• Browsing and computer use. Claude can open sites, read pages, click and fill forms through the built-in browser or Claude in Chrome. Computer use, in which Claude operates apps on your screen directly, is in beta on Pro and Max and is the capability Anthropic itself asks users to treat most cautiously.
• Scheduling, memory and projects. Any task can be put on a daily, weekly or monthly schedule. In cloud sessions Claude starts from what it remembers from your chats and carries what it learns back. Projects group related tasks into workspaces with their own files, instructions and memory.

Anthropic also ships Claude for Small Business, announced on 13 May 2026: a toggle inside Cowork that bundles connectors for QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365 with fifteen ready-made workflows such as a month-end close, an invoice chaser and a lead triager. It is framed for the US market and built around QuickBooks and PayPal, so for a UK firm on Xero the relevant parts are the HubSpot, Workspace and 365 connectors and the workflow patterns, which transfer directly.

Which plan a UK firm needs, and what it costs

Cowork is not sold separately. It is included in every paid Claude plan: Pro, Max, Team and Enterprise. For a business the question is therefore the plan question, and for a firm of two to 150 people the answer is almost always Team, which adds central billing, single sign-on, admin controls over connectors, usage analytics, enterprise deployment of the desktop app and no model training on your content by default. Enterprise is for firms that need audit logs, SCIM provisioning, the Compliance API, custom data retention, role-based access or a HIPAA-ready configuration, and it carries a 20 seat minimum and annual billing. Our guide to Claude Team versus Enterprise works through the line that forces the upgrade.

On the prices, a caution first. On 24 October 2026 we read claude.com/pricing from a UK browser and it showed US dollar prices with the line "Prices shown don't include applicable tax". On 15 September the same page showed a UK visitor sterling prices with the line "Prices include 20% VAT", which is what our sterling cost guide records. The display has changed between visits this year, so read the figure on the page when you buy and check whether VAT is inside it. The dollar figures shown on 24 October, with an indicative conversion at HMRC's October 2026 rate of USD 1.3456 to GBP 1, were:

Plan	Price shown (USD)	Indicative GBP before VAT	Notes
Pro	$17 a month billed annually, $20 monthly	About £12.63 or £14.86	Individual plan; Cowork included
Max	From $100 a month	About £74.32	5x or 20x Pro usage; monthly only
Team standard seat	$20 per seat per month billed annually, $25 monthly	About £14.86 or £18.58	2 to 150 seats; more usage than Pro
Team premium seat	$100 per seat per month billed annually, $125 monthly	About £74.32 or £92.90	5x standard usage; seat types can be mixed
Enterprise	$20 per seat per month billed annually plus usage at API rates	About £14.86 per seat plus usage	20 seat minimum; annual only

The figure that catches firms out is not the seat price but usage. Anthropic states plainly that Cowork consumes limits faster than chat, because every step of a multi-step task uses tokens: the planning, each file read, each tool call, the code it runs. Limits reset on a rolling five-hour window with weekly caps on top, and all of a person's Claude activity draws from one pool. A standard Team seat is enough for someone who hands Cowork two or three substantial tasks a week. Someone who runs it all day needs a premium seat, or the plan's usage credits turned on so they can keep working at API rates once the limit is hit. Budget for a mix, not a single seat type, and read the usage analytics after the first month before fixing the ratio.

Ten jobs to hand Cowork first

The right first jobs share three properties: the inputs already exist as files or in a connected system, the output is a document someone currently assembles by hand, and a mistake is visible before it costs anything. These ten meet that test for most UK firms. Each is a pattern Anthropic documents or demonstrates; the UK framing is ours.

1. The weekly management report, on a schedule. Point Cowork at the folder or drive where the week's figures land, specify the slides and the thresholds that should be flagged, and schedule it for Monday at 8am. Anthropic's own example does exactly this from Amplitude and a Drive tracker. The report is ready for review when the meeting starts instead of being assembled in it.
2. Month-end reconciliation from exports. Give it the folder of bank and ledger exports and the budget file, and ask for a working spreadsheet with variances flagged and a note on what did not reconcile. For a Xero firm the exports come from Xero; the connector can pull them directly once its scopes are set.
3. First-pass contract review against your playbook. A folder of inbound agreements plus a document setting out your standard positions produces a memo per contract listing each departure, cited to the clause, with the gaps your playbook does not cover marked for a human. Anthropic demonstrates this pattern; for a UK firm the playbook is the work, and our law-firm guide covers the confidentiality controls around it.
4. Meeting notes and transcripts into actions. Drop a quarter of call notes in a folder and ask for themes, open commitments and recurring objections, each quoted from the note it came from. Sales teams get a read on every open deal; operations teams get a decisions log.
5. Inbox and channel summaries. With Gmail or Microsoft 365 and Slack connected, a scheduled morning summary of what arrived overnight, what needs a reply and what was decided in the team channels. Keep this read-only; the point is the summary, not the replies.
6. Tender and bid assembly. Point it at the invitation to tender, your library of previous responses and your case studies, and ask for a first draft mapped question by question with the gaps listed. The gaps list is usually the most useful output.
7. Policy and handbook drafting from templates. HR and operations policies drafted from your existing documents and a short brief, in your house style, as editable Word files. Nothing goes to staff without a human sign-off, but the blank page is gone.
8. Supplier and quote comparison. A folder of PDFs from suppliers becomes a comparison spreadsheet with like-for-like columns, the exclusions each quote carries and the questions to ask before choosing.
9. CRM hygiene. With HubSpot connected, a monthly task that lists stale deals, contacts missing an owner or a next step, and duplicates, as a report for the sales lead to act on. Leave the writing back to the CRM on approval until the report has proved itself.
10. Receipts and expenses. A shared folder that staff drop receipts into, turned into a formatted expense report with totals by category on a schedule. Small, boring, and the kind of job that demonstrates the pattern to a sceptical team.

Notice what is not on the list: anything that sends, pays, posts or deletes without a person approving it, and anything that touches payroll, bank access or client identity documents. Those come later, if at all, once the read-only patterns have earned trust.

The settings to change before anyone starts

Cowork on a Team plan arrives with defaults chosen for speed. Anthropic documents each of them, and for a business most of them should be reviewed by an Owner before the first seat is handed out. In Organisation settings:

• Cowork itself is on by default. The toggle to disable it exists if you want to stage the rollout by team; on Enterprise you can use groups and custom roles to enable it for specific teams instead.
• Sessions in the cloud are on by default on Team and off by default on Enterprise. Cloud sessions are what give you work that continues when the laptop closes; they are also what sends local files through Anthropic's servers. Decide deliberately, and record the decision in your DPIA.
• Automatically approve mode. Two settings: whether members may use it at all (on by default) and whether new tasks start in it (on by default on Pro, Max and Team, off on Enterprise). In auto mode Claude reviews each action for safety before it runs and blocks anything it judges unsafe; in manual mode it asks you first. For a first month in a business, start new tasks in manual.
• Always allow for connector tools is off by default and should stay off. It is the setting that would let a member skip per-task approval for write-capable connector tools, the ones that send, change or delete things in Xero, HubSpot or your inbox. Read-only tools are exempt only where the connector marks them as such, and most custom connectors do not.
• Browser and web search. The built-in browser is on by default on Team. Web search can be turned off for Cowork and chat. Decide which sites Claude may work in; Anthropic's advice is to limit browsing to trusted sources and never to use the browser for anything that handles money or personal information.
• Monitoring. Cloud sessions are captured in the Compliance API, and Team and Enterprise owners can stream Cowork events to a SIEM through OpenTelemetry. Local session history, by contrast, is stored on each user's computer, is not subject to Anthropic's retention policies and cannot be centrally deleted. Enterprise administrators can retrieve it through the Compliance API for the most recent 30 days. Your retention schedule needs to say what happens to it.

Behind the settings sits the risk Anthropic names most often: prompt injection. Cowork reads content you did not write, in web pages, emails and documents, and that content can contain instructions aimed at Claude. Anthropic trains against it, scans untrusted content and screens actions in auto mode, and still says the risk is non-zero. Its own mitigations are the ones above: limit what Claude can read, limit what it can do, keep write actions on approval, and stop any task that starts touching files or sites you did not mention. For the UK data-protection side, our guides to Claude's data residency, whether Claude is GDPR compliant and running a DPIA for an AI assistant rollout cover lawful basis, transfers and retention.

Cowork, chat, Claude Code and Projects: who gets what

A sensible allocation for a non-technical firm is simple. Everyone gets chat (or, on the unified experience, simply Claude) for everyday drafting, questions and analysis. Everyone who has a recurring, file-based job gets Cowork for it, starting with the ten above. One person, whoever maintains the shared setup, uses Claude Code to keep the organisation instructions, skills and connector configuration in a folder with version history. Projects hold the context for each recurring job: the brief, the template, the house style, the files. Skills capture how a task should be done so it is done the same way every time. Our comparison of Projects, Skills, Code and Cowork goes surface by surface, and the guide to Claude Code and Cowork for non-technical teams covers the seat maths and the handover pack.

Four mistakes to avoid

• Treating it as a faster chat. Cowork rewards a brief written the way you would brief a colleague: the outcome, the format, and the inputs it should work from. Anthropic's own getting-started advice says exactly that. One-line prompts produce one-line results and burn usage doing it.
• Leaving auto-approve as the default for everyone on day one. It is the right setting for well-defined, read-mostly work once it has earned trust. It is the wrong setting for a team's first week with connectors to the finance system.
• Scheduling consequential tasks. Anthropic's guidance is unambiguous: start scheduled tasks with summaries, do not schedule anything that sends messages, pays or deletes, review every run, pause what you are not using. A scheduled task runs while nobody is watching.
• Forgetting the local history. If any sessions run locally, there is conversation history on laptops that your retention policy does not reach through Anthropic. Decide what to do about it before, not after, a subject access request.

Where The AI Consultancy fits

Most of the value in Cowork comes from three things nobody sells with the seat: the brief, the connector scopes and the settings above. Our Claude Enablement service installs and configures a Team or Business workspace on site at a fixed price (Team Install from GBP 4,500 ex VAT, Business Install from GBP 7,500 ex VAT), sets those defaults, writes the first skills and projects with your team, and includes a 30-day adoption window. Where the question is still which plan, which model and whether the governance case holds, our Claude consulting service answers it first. We are Certified in the Claude Partner Network and take no referral or reseller income from Anthropic, as our independence page sets out.

Sources

• Claude Help Center, "Get started with Claude Cowork", "Use Claude Cowork on Team and Enterprise plans", "Use Claude Cowork safely" and "Claude Cowork and chat are one Claude". Accessed 24 October 2026.
• Claude Help Center, "Release notes": Cowork generally available on desktop 9 April 2026; web and mobile 7 July 2026; unified experience 16 September 2026. Accessed 24 October 2026.
• claude.com/product/cowork and claude.com/pricing, read from a UK browser on 24 October 2026; both showed US dollar prices with the line "Prices shown don't include applicable tax".
• HMRC monthly exchange rates, October 2026: USD 1.3456 per GBP 1 (trade-tariff.service.gov.uk).
• Anthropic, "Introducing Claude for Small Business", 13 May 2026.

This guide is general information, not legal, data-protection or procurement advice. Anthropic's prices, plan features and defaults change; confirm them on its pages before buying or configuring.