If you're paying for Microsoft 365 or Google Workspace on a business plan, AI has already been quietly showing up in your apps for a while. Copilot in Outlook can summarise email threads and draft replies. Gemini in Docs can help you write and refine content. These are genuinely useful features, but they're largely reactive; you ask, they respond, you do the rest.
With the advent of two new tools, that's about to change: Claude Cowork from Anthropic and Copilot Cowork from Microsoft. Rather than helping you with a task while you're in it, they can take the task off your hands (for the most part - I'll go into a bit more detail later on).
I've been experimenting with Claude Cowork the last week or so, and I wanted to share what I found; including some honest caveats about privacy that are worth knowing before you dive in.
Both Claude Cowork and Copilot Cowork are what's called agentic AI; a tool that can take actions, not just give advice. Rather than answering questions and leaving the legwork to you, these tools can actually do things on your behalf, such as accessing your local files, your calendar, your email, and other applications or services you might use for your business.
Importantly, neither of them charges ahead without your approval. Both generate a plan first and show you what they intend to do before touching anything. You can review, adjust, or cancel. That "human-in-the-loop" design matters: there's a genuine fear that an AI acting without oversight could result in some unexpected behaviours.
As an aside, this kind of anxiety isn't new; as the "infrastructure as code" (IaC) paradigm started in the naughties, there was a running joke that automating things meant that mistakes could propagate much faster (including the destruction of your cloud-hosted infrastructure)! Despite this, the benefits of being able to create infrastructure to an exact specification in minutes outweighed the potential risk, and the industry evolved to create best practices and tooling that mitigated that risk. I think we are seeing that kind of maturity in the Cowork tools already.
My downloads folder has, quite frankly, always been a mess. It accumulates cruft over months (years?) of documents, installers, scripts and ISO files being grabbed to support the various projects I've worked on. Sorting it out manually never quite makes it to the top of the to-do list, and so I pointed Claude Cowork at it and asked for help.
What impressed me wasn't just that it tidied things up, but how it went about it. Before moving a single file, it identified files that were already present in my Documents directory structure and suggested their deletion from the Downloads directory. For everything else, it categorised the files, giving me suggested actions for each. It flagged things I'd forgotten were there such as documents that probably needed attention rather than deletion. Once I'd reviewed the suggestions and approved, it moved important files sensibly into my existing Documents directory structure and put together a script to clean up what was left.
It was genuinely useful: this is the sort of task that would have taken me the best part of an hour to do myself, requiring tedious clicking, reviewing and moving. There's a good chance if I had decided to do this myself, I would have eventually just deleted all the contents of the Downloads directory! It took Claude a few minutes, handling common office formats (Word documents, spreadsheets, PDFs, presentations without any issues. Most importantly, I could just leave it running during that initial phase of discovery and get on with other work.
You can also schedule tasks to run automatically; daily, weekly, monthly. For things like generating a weekly summary from project files or keeping a folder tidy, that's where it starts to feel less like a tool and more like genuine assistance.
Other tasks I'll be trying over the coming weeks:
Finally, Anthropic have started publishing plugins, which can provide specific context/skills to Claude; it's all very exciting and could provide business owner with a tonne of value.
There are two separate products here, and it's worth being clear about the difference.
Claude Cowork lives inside the Claude desktop app, which runs on Mac and Windows (there's also an unofficial Linux desktop client). It works across your local files and folders, and connects to third-party tools via a growing library of integrations; Google Drive, Gmail, Google Calendar, DocuSign, and others. It can also link Excel and PowerPoint workflows together, so data can flow between them without you manually copying anything across.
One notable point: conversation history is stored locally on your device, not on Anthropic's servers. That's a reasonable privacy protection for the chat log, but it doesn't mean your data stays on your machine during use. More on that below.
Claude Cowork is available on paid Claude plans (Pro, Max, Team, Enterprise). It's currently in research preview, meaning it's functional and worth trying, but expect continued development.
If you're already using Claude, run Linux or macOS, or your business isn't heavily embedded in Microsoft 365, Claude Cowork is the natural starting point.
Copilot Cowork is Microsoft's equivalent, built directly into Microsoft 365. It can take action across Outlook, Teams, Excel, Word, PowerPoint and SharePoint; all the tools that many businesses already live inside every day.
A note on plans: to get Copilot embedded in the Microsoft 365 desktop apps (Word, Outlook, Excel and so on), you need a qualifying Microsoft 365 Business subscription plus the Microsoft 365 Copilot add-on, which runs to around £20 per user per month on top of your existing licence. It's not included in the base plans (the Basic, Standard or Premium tiers, for example). Similarly, for Google Workspace users, Gemini features in Gmail, Docs and Sheets are included in Business Standard and above; but the more advanced capabilities require the AI Expanded Access add-on from March 2026. The point being: if you're already using these platforms and wondering why you don't see AI features, it's likely a plan/licencing question worth checking.
Interestingly, Microsoft has integrated the technology behind Claude into Copilot; so the two products aren't entirely separate under the hood.
|
⚠️ Worth Knowing At the time of writing, Copilot Cowork is not yet generally available. It's in research preview with a limited set of customers, with broader rollout through Microsoft's Frontier programme expected in late March 2026. If you're a Microsoft 365 shop, it's very much one to watch; but you can't go and try it today the way you can with Claude Cowork. |
This is the bit people tend to skip (or simply not be aware of), and it's worth a few minutes of your time.
When Claude Cowork accesses a file and reasons about it, the content of that file is sent to Anthropic's servers to be processed by the model. That happens regardless of any other privacy settings. The 'stored locally' feature refers to your conversation history; not to whether your file contents are transmitted. These are two separate things, and it's important not to conflate them.
This doesn't mean you shouldn't use it; but it does mean you should be thoughtful about what you point it at, particularly if you're handling client data or anything commercially sensitive.
If you're using a personal Claude plan (Free, Pro, or Max), Anthropic may use your conversations to improve future versions of the model, unless you opt out. The good news is that opting out is straightforward:
For an extra layer of protection, use incognito mode for any session involving sensitive material; incognito chats are never used for training, regardless of your account settings
|
⚠️ Worth Knowing Opting out stops future use of your data, but it won't undo anything already included in a training run that's in progress. It's prospective, not retroactive. |
Claude Team and Enterprise plans operate under commercial terms, which means your data is not used for model training at all. If you're planning to use Cowork regularly with business data, or if you operate in a regulated sector (financial services, legal, healthcare), a Team or Enterprise plan is the right starting point, not an upgrade to consider later.
On data residency, this is an area where the current answer is limited. Anthropic does offer a Zero Data Retention option for enterprise deployments, and data residency controls are available when Claude is accessed via AWS Bedrock or Google Vertex AI. Unfortunately for standard Claude users, there are no granular data residency controls. If your industry requires data to remain within specific geographic boundaries, this is a conversation to have before adopting the tool, not after.
Yes: with sensible expectations.
These tools are genuinely useful, and the best use cases right now are the things you find yourself putting off: the messy folder you haven't touched in six months, the meeting you always walk into underprepared, the weekly roundup you mean to send but never quite get around to.
They're not magic, and they're not finished products. Both are in research preview, which means they're capable but still evolving. The right approach is to start with low-stakes tasks using non-sensitive data, build a feel for how they work, and expand from there.
As a rough guide:
The direction of travel is clear. AI is moving out of the chat window and into the workflow. For small businesses, that's a genuine opportunity to get time back on the kind of tasks that eat your day without adding much to your bottom line.
This is exactly the kind of thing we help businesses think through; whether a tool is appropriate, how to trial it safely, and how to make it part of a sensible workflow rather than just another thing to manage. If you'd like to have that conversation, book a call. No obligation, just a practical chat.