Microsoft Office just became a Cowork surface
The real shift isn’t Claude inside Excel. It’s what happens when your spreadsheet, memo, deck, and inbox start sharing context.
Claude now works across Microsoft 365 apps.
That headline will probably get squeezed into the usual fight.
Claude vs Copilot.
Anthropic vs Microsoft.
Another sidebar inside another office app.
That framing misses the useful part.
Claude can coordinate between Excel, PowerPoint, Word, and Outlook add-ins. Anthropic’s docs say Claude can read from one Microsoft app and make changes in another, such as analyzing an Excel workbook and creating a PowerPoint presentation from the results without manual copy-paste.
That matters because office work rarely breaks inside one file.
It breaks while moving between files.
A spreadsheet holds the numbers.
The memo explains what they mean.
A deck turns the explanation into something leadership can skim.
Outlook carries the original ask, the deadline, the attachments, the politics, and the thread nobody wants to reread.
Most people call that “office work.”
Operators know it’s a handoff problem.
Someone pulls numbers from Excel. Another person explains them in Word. A deck gets assembled from the memo. Then the final email turns the whole thing into a decision request.
That chain breaks constantly.
A stakeholder asks why the deck says pipeline improved, the memo says pipeline was flat, and the spreadsheet shows the movement came from one late-stage account.
Now the team isn’t making a decision.
They’re reconciling versions.
Claude’s Microsoft 365 support is interesting because it touches that ugly middle layer of work. The part where context gets rebuilt, compressed, distorted, or lost.
That’s where Claude Cowork starts to matter.
Treat this like a packet problem
A beginner doesn’t need to understand Microsoft Graph, MCP, cloud gateways, token storage, or admin consent to get the first useful idea.
Start with the packet.
A packet is a reviewable bundle of work.
It could be a weekly status memo, board update, client prep brief, customer escalation summary, legal consistency check, PowerPoint outline, or hiring packet.
The format changes, but the job stays similar.
You’re taking scattered material and turning it into something a person can review, edit, approve, or send.
That’s the sweet spot.
Claude doesn’t need to “run the business” for this to be useful. It needs to keep the source material, output format, and review step connected long enough to reduce manual stitching.
This is exactly where Microsoft 365 work fits when it’s scoped properly.
Useful Cowork work tends to be multi-step, context-heavy, deliverable-oriented, reviewable, recurring, and still human-owned when judgment matters.
That sounds boring.
It’s also where the money usually hides.
What Anthropic’s docs actually say
The current cross-app support has a few practical requirements.
You need a paid Claude plan. You need the Excel, PowerPoint, Word, and Outlook add-ins installed from the Microsoft Marketplace. Team and Enterprise users may need an organization owner to enable the “Let Claude work across apps” setting before individuals can turn it on. Anthropic says the setting is default on for Pro and Max, and default off for Team and Enterprise.
That detail matters.
This isn’t just a feature toggle for one person in a vacuum.
On Team and Enterprise plans, someone has to decide whether this should be enabled across the organization.
Once it’s active, Claude can use the Microsoft 365 add-ins to read from and write to open files. Context transfers between apps automatically, so users don’t need to copy and paste the same information between Excel, PowerPoint, Word, and Outlook.
There are limits.
Claude can only read from and write to files that are currently open in the Microsoft apps. It can’t create, open, close, or switch files directly from the add-ins. Cross-app chat history isn’t saved between sessions.
That should change how people use it.
Don’t imagine Claude wandering through your entire Microsoft tenant from the add-in pane.
Think of it as a workbench.
You open the files, define the packet, let Claude help move the job across those files, then review what changed before the work leaves your desk.
That’s enough to be valuable.
It’s also enough to create a mess if the wrong files are open or the task is vague.
The connector is a different layer
The Microsoft 365 connector and the Microsoft 365 add-ins are related, but they aren’t the same thing.
The connector lets Claude search, analyze, and access information across SharePoint, OneDrive, Outlook, and Teams. Anthropic says it’s available on all Claude plans, but it requires a Microsoft Entra tenant tied to a Microsoft Business plan. Personal Outlook, Hotmail, and Live accounts can’t be used with it.
The add-ins put Claude inside the Microsoft apps where the work gets shaped.
A normal user can think about it this way:
The connector helps Claude find context.
The add-ins help Claude work with open files.
A consultant preparing for a client review might use Microsoft 365 context from SharePoint or Outlook to understand the account history, then use Word and PowerPoint add-ins to create a prep memo and deck outline.
A founder might use Outlook and Teams context to understand what a customer asked for, then use Excel and Word to create a decision memo.
An operator might take a weekly workbook, a notes document, and an existing presentation, then turn them into a leadership update.
More access isn’t automatically better.
The right access reduces repeated handoff work when the output is clear and reviewable.
Real users are already showing the demand
The strongest signal isn’t the launch copy.
It’s what people are trying to do with the tools.
One Reddit user asked for help with many interconnected Office 365 Excel files containing sensitive data. Their problem wasn’t “how do I write a prompt?” It was the practical mess of using Claude Cowork with large Office files while sensitive information was still inside them.
Another Reddit user described Claude for Excel working through complex financial models with logic spread across 10 sheets, circular references, formulas, dependencies, and buried mistakes. Their underlying use case was concrete: model inspection, dependency tracing, and mistake detection inside messy workbooks.
A different Reddit user described Claude’s Word add-in across dense legal documents, each 40, 60, or 100+ pages, while also pulling from a spreadsheet workbook with 10 worksheets. The value wasn’t “write me a paragraph.” It was consistency across a package of documents.
A Hacker News thread on Anthropic’s finance-agent push made the same point from another angle. One commenter said a lot of financial-services work is slides and Excel documents, not AI moving money around. Another pushed back that domain knowledge in finance often lives in conversations and judgment, not just documents.
Both sides are useful.
One side shows why Microsoft 365 workflows matter. Excel, slides, memos, and deal materials make up a huge surface area of business work.
The other side keeps the hype in check. Claude can help shape the packet. It can’t magically know every relationship, backchannel, client nuance, or unwritten assumption.
That’s the right tension.
This doesn’t replace judgment.
It makes the material easier to inspect before judgment happens.
The beginner-safe workflow
The best first workflow is a numbers-to-memo-to-deck packet.
It’s easy to understand, tests the cross-app feature directly, and applies to founders, operators, analysts, finance people, consultants, and anyone who has to turn data into an explanation.
Use non-sensitive material the first time.
Open one Excel workbook, a blank Word document, and a blank or template PowerPoint file.
Keep Outlook closed until the memo and deck outline have been reviewed.
Then ask Claude to inspect the workbook, identify the important changes, draft a plain-English memo, and convert that memo into slide-ready bullets.
Before any email gets drafted, ask Claude to show what it carried forward from the workbook into the memo and from the memo into the deck.
Most people will skip that last step.
Don’t.
If Claude is moving context across apps, you need to know what moved.
How to run this if you’re not technical
Pick one workbook you understand.
Don’t start with your entire OneDrive.
Avoid payroll, contracts, board materials, customer health records, private employee data, or anything regulated.
Use a low-risk report first.
Your goal is to learn the workflow, not test the worst possible edge case on day one.
Open the workbook.
Prepare the Word document where the memo should go.
Use a PowerPoint file or template that’s safe for the session.
Turn on the cross-app setting if your plan and workspace allow it.
Then give Claude a job with a clear ending.
Don’t write:
“Analyze this.”
Use something closer to:
“I need a reviewable leadership packet from this workbook. Inspect the workbook, explain the major changes in plain English, draft a Word memo, then prepare a PowerPoint outline. Before anything moves into the deck, show me which facts and assumptions you’re carrying forward.”
That wording does several useful things.
It names the output, limits the job, separates facts from assumptions, creates a review point, and avoids pretending the result is final.
That’s how a beginner gets value without becoming the cleanup layer for a vague automation.
What power users should notice
Advanced users should care less about the sidebar and more about the boundary.
Anthropic’s Microsoft 365 connector security guide says the connector requires a Microsoft Entra tenant tied to a Microsoft Business plan. It also says personal Microsoft accounts can’t be used, and that Graph API calls made by the connector are logged in the organization’s Microsoft 365 audit log.
That’s useful.
It doesn’t remove the need for workflow design.
Read access still matters. A model doesn’t need write access to leak sensitive context into the wrong draft. If the wrong information gets summarized into a memo or copied into a deck, the damage can happen before anything is formally sent.
Anthropic also says Cowork can access files, browser activity, connected services, and apps, and that users shouldn’t use Cowork for regulated workloads. Cowork activity isn’t captured in audit logs, the Compliance API, or data exports.
A technical operator still needs answers to several plain questions.
Who can enable this?
Which connector tools are available?
Where do SharePoint and OneDrive boundaries sit?
What Outlook content can be surfaced?
Which add-ins are deployed?
Can apps pass context to each other?
What review step happens before outputs leave the draft layer?
That’s the difference between “Claude can access Microsoft 365” and “we’ve designed a workflow we can trust.”
The cross-app safety issue
The risk isn’t only that Claude might write the wrong sentence.
The sharper risk is that the wrong context travels into the wrong output.
Anthropic’s Microsoft 365 docs say context transfers between apps automatically and that Claude carries relevant context forward while working across Excel, PowerPoint, Word, and Outlook. The same page says add-in activity isn’t currently included in Enterprise audit logs, the Compliance API, or data exports.
That’s not a small caveat.
It’s the operating rule.
Cross-app context is useful because work stops getting stranded in one app.
It’s risky because sensitive information can follow the work farther than the user intended.
A safe beginner rule:
Only use cross-app workflows with files you’d be comfortable seeing in the final packet.
A better operator rule:
Classify context before it moves.
Use three labels.
Keep: approved, relevant, and safe to include.
Check: useful enough to consider, but it needs a human decision.
Block: sensitive, stale, private, unsupported, or not allowed to leave the source app.
That small review habit is more useful than another clever prompt.
The audit gap should change the use case
Claude Cowork activity isn’t captured in Audit Logs, the Compliance API, or Data Exports. Anthropic says not to use Cowork for regulated workloads.
The Microsoft 365 cross-app docs make a similar point for the add-ins. Inputs and outputs are deleted from Anthropic’s backend within 30 days except as described in Anthropic’s retention terms, but the add-ins don’t inherit custom data retention settings. Add-in activity also isn’t currently included in Enterprise audit logs, the Compliance API, or data exports.
That means the tool is strongest for reviewable drafting, internal packets, working summaries, and controlled prep work.
It’s a weak fit for workflows where the organization needs complete audit coverage.
That doesn’t make it useless.
It tells you where not to force it.
A team that needs strict auditability should treat this as a drafting and prep surface, not the final governed system of record.
The third-party platform detail advanced teams shouldn’t miss
Some enterprise users won’t route prompts through a normal Claude account. They may use a gateway, Amazon Bedrock, Google Vertex AI, or Azure AI Foundry.
Anthropic’s third-party platform docs say end users can connect the Microsoft add-ins through an enterprise gateway. Credentials are stored locally in browser localStorage inside the add-in’s sandboxed iframe, and Anthropic warns users to enter gateway-issued tokens instead of raw cloud provider credentials.
That one sentence deserves attention.
If your team uses a gateway, the add-in becomes part of your internal model-routing and credential story.
You need answers on token issuing, gateway scope, allowed models, add-in access, CORS rules, token rotation, and offboarding.
None of that belongs in a beginner tutorial.
It absolutely belongs in the mind of the person deploying this across a real team.
The beginner sees a sidebar.
The operator sees a trust boundary.
What to automate first
Don’t start with the most dramatic workflow.
Start with a boring packet that repeats.
Strong candidates happen often enough to matter, use source material a human understands, end in a known output format, allow review before use, and carry manageable downside if something needs correction.
Good first workflows include weekly metrics memos, client prep packets, internal meeting briefs, PowerPoint outlines from reviewed Word memos, consistency checks across related documents, and customer escalation summaries from known threads.
Riskier first workflows include legal advice, payroll review, compliance reporting, high-stakes financial decisions, unsupervised customer replies, or anything involving medical, financial, legal, or private employee data.
The difference isn’t sophistication.
It’s reviewability.
A boring workflow with a review point beats a flashy one that hides the failure until after the output is sent.
What this unlocks for different roles
Founders can use this to turn scattered context into a decision packet. The packet still needs judgment, but the prep work gets less chaotic.
Operators can use it to reduce the weekly copy-paste loop across files, emails, and slides.
Analysts can turn spreadsheet findings into a clear memo while preserving caveats around the numbers.
Consultants can move client prep from notes, account history, and spreadsheets into one reviewable brief.
Marketers can pull campaign data, positioning notes, and a prior deck into a messaging update without restarting the brief.
Recruiters can organize role notes and interview feedback into a structured packet while keeping final hiring judgment human-owned.
The pattern is consistent.
Claude gathers, transforms, and packages.
The human checks meaning, sensitivity, tone, and consequence.
That split is the whole game for Microsoft 365 workflows.
Where the workflow can break
A few failures are predictable.
Claude might summarize too aggressively and drop the caveat that mattered.
Draft assumptions can end up inside final-looking slides.
Weak trends may sound stronger than the spreadsheet supports.
An open file you forgot about can feed context into the session.
A polished email can make an unfinished memo feel ready.
The formatting itself can create false confidence.
A tidy memo can make a weak claim feel approved.
A nice deck can make an assumption look like a conclusion.
A fluent email can make missing context sound intentional.
The fix isn’t paranoia.
Ask Claude to show the handoff.
Between Excel and Word, check which facts and assumptions moved forward.
When the memo becomes slides, review what got compressed.
Before Outlook enters the workflow, force a final approval pass.
The operator workflow
Here’s the version I’d teach first.
Open only the files needed for the packet.
Use cross-app work when those files are safe for the session.
Ask Claude to create a handoff map before drafting.
Approve the source apps, output format, and sensitive-data rules.
Let Claude inspect the workbook or document.
Review the memo for facts, assumptions, and missing caveats.
Move into the slide outline after the memo is approved.
Request a context carryover log.
Draft the email last.
Outlook is where the packet leaves the room.
Treat it that way.
The larger opportunity
Microsoft 365 support could turn Claude into something more useful than another chat pane.
But that only happens when users stop treating Office apps like isolated containers.
The useful workflow isn’t “Claude in Excel.”
It’s source material to memo, memo to deck, deck to reviewed message, and reviewed message to human-approved send.
That is a workflow operators already understand.
Claude becomes useful when it reduces the manual stitching between those steps without hiding what changed.
The crowded angle is “AI inside Office.”
The better angle is “office work finally gets a handoff layer.”
That’s why this matters for Claude Cowork.
Not because Claude can appear in another sidebar.
Because the work can keep its shape across more of the places where business already happens.
