Free resource · Word + Excel kit
Alkymia di Grandha's Cowork : delegating real work to Claude Cowork and to Copilot Cowork
Claude Cowork and Copilot Cowork They no longer simply answer a question: they carry out a complete task, using your files and applications, right up to the deliverable. This kit provides employees, managers, and sponsors of large organizations with the briefing methodology, 24 tasks to delegate, safeguards, and a 30-day pilot plan, with seven ready-to-use tools.
The kit Cowork, in short
- What it is
- A kit for transitioning from chat to coworking: entrusting a complete, multi-step task, including files and applications, to an AI, and receiving a finished deliverable for review. He compares Claude Cowork and Copilot Cowork as they are in October 2026, gives a briefing method, 24 missions to delegate, safeguards, the position of the French authorities and a 30-day start-up plan.
- For who
- Employees who want to delegate real work to AI, managers who want to make it a team practice, and sponsors (business management, CIO, CISO) who have to decide how to open up these tools in a large organization.
- Format
- A 17-page Word document in 11 chapters and an Excel workbook with 7 tools: a library of 24 missions with ready-to-copy briefs, a brief generator in five blocks, a traffic light system "Can I delegate?", and a choice between Claude Cowork and Copilot Cowork by weighted criteria, sponsor deployment checklist, delegation log and time saved, 30-day pilot plan.
- What it contains
- What's the difference between a chat, a coworking space, and a team agent? Claude Cowork in October 2026 (operations, administrative controls, announced merger with the chat); Copilot Cowork in October 2026 (validations, skills, license, credits, Anthropic models in Europe); how to choose; the brief method; 24 missions per function; the safeguards and prompt injection; what CERT-FR, CNIL and the AI Act say; measuring time saved; the pilot plan; resources.
- What is it based on?
- The official documentation from Anthropic and Microsoft (release notes, security guides, FAQs, and Copilot governance) Cowork, documentation on Anthropic models in Microsoft 365 Copilot and on Microsoft Purview), the CERT-FR bulletin CERTFR-2026-ACT-016, the CNIL and Council for AI and Digital Technology note on agentic AI, the AI Act, the Epoch AI and Ipsos study on delegating work to AI, and our workshops in large organizations.
- Price
- Free, in exchange for a professional email address. No unsolicited emails.
- Update
- Edition October 2026, information verified on October 12, 2026. These products evolve every week: check the license and configuration points before making any decision.
- What it doesn't replace
- Validation from your CIO, CISO, and DPO, nor legal advice on your situation, is required. The kit provides the methodology and key questions to address; the decision to open the tool remains yours.
- Go Further
- examples of missions Claude Cowork by job, Claude comparison CoworkCopilot Cowork and ChatGPT Work, guide to setting up Claude Cowork, diagnostic Cowork, Security and data kit, AI charter template.
With Claude Cowork and Copilot CoworkGenerative AI is changing its nature: instead of being asked to provide an answer to be copied elsewhere, it is given a task. The agent establishes a plan, opens authorized files and applications, progresses in several stages, returns to you at key moments, and delivers a document, spreadsheet, or presentation ready for review. Anthropic analyzed 1,2 million Claude sessions. Cowork Between May 11 and 31, 2026: one third of uses relate to business processes and operations, and about half relates to "work around work": gathering, sorting, formatting, relaunching.
The gain is not automatic, however. An Epoch AI and Ipsos study published in August 2026 shows that approximately one in six assisted tasks takes longer than before, and that more than one in four deliverables requires significant revisions. This toolkit translates coworking into concrete reflexes for a large organization: choosing between Claude Cowork and Copilot Cowork, write a brief that avoids vague deliverables, know what can be delegated and what never is delegated, take into account the positions of CERT-FR and CNIL, measure the actual time saved and launch a supervised pilot.
1. From chat to coworking: what's changing
Microsoft summarizes Copilot Cowork Thus, the tool transforms a request into a plan and executes it across your applications and files. Anthropic describes the same process for Claude: it's no longer about discussing with the AI but about assigning it a task, which it carries out while you do something else. The toolkit distinguishes three ways of working with AI, each with its own set of rules and risks.
| Chat | Cowork | Team Agent | |
|---|---|---|---|
| Examples | Copilot Chat, ChatGPT, Claude in conversation | Claude CoworkCopilot Cowork | Agent Builder, Copilot Studio, agents Claude |
| What is entrusted to him | A question, a text to rephrase | A multi-step task, lasting from a few minutes to a few hours. | A recurring process shared by a team |
| What he gives | A copyable response | A finished deliverable: file, spreadsheet, drafts, summary | A service that is always available |
| Who builds it | Nobody uses it | The user, for himself | A designated contact person or IT department, with a named owner |
| Main risk | Error in the text | Actions on files, emails, and calendars | Large-scale repeated error |
The simple rule: a personal routine that you delegate falls under coworking; a team process, equipped with tools, access rights, and an owner, falls under an agent built in Copilot Studio or Agent Builder. Many team agents, in fact, originate from a coworking setup that has proven successful for a single individual.
2. Claude Cowork or Copilot Cowork Where do they stand in October 2026?
Claude Cowork is generally available on macOS and Windows since April 9, 2026, and on the web and mobile since July 7. On September 16, 2026, Anthropic announced the merger of Cowork and the cat in a single experience "Claude": the abilities Cowork become available from any conversation. Copilot CoworkMicrosoft 365 Copilot's agentic capability has been generally available since June 16, 2026; according to Microsoft, more than half of the Fortune 500 companies were using it during the Frontier phase. Both tools address the same need, but in different ways.
| Criterion | Claude Cowork | Copilot Cowork |
|---|---|---|
| Playground | Local files and connected folders, connectors to numerous applications, web and mobile | Microsoft 365 only: Outlook, Teams, OneDrive, SharePoint, Office |
| Default access | Included in the paid plans, the administrator can disable it. | Closed until the administrator opens a spending policy |
| Economic model | Subscription per seat, with usage quotas consumed more quickly than in chat | Microsoft 365 Copilot license, plus Copilot Credits billed on a usage basis |
| Models | Claude | Several models; Claude is disabled by default in Europe |
| Action control | Request before action by default, authorization before any permanent deletion | Validation before sending, publishing, meeting or modifying the file |
| Governance | Compliance API, OpenTelemetry, role-based and group-based control | Microsoft Purview, unified audit log, spending limits |
| Ideal for | File analysis, production of rich deliverables, multi-tool teams | "All Microsoft" organizations, tasks involving email, meetings and documents |
One point is particularly important for a French organization: in the European Union, EFTA, and the United Kingdom, Anthropic templates are disabled by default in Microsoft 365 Copilot and excluded from the EU Data Boundary, according to Microsoft documentation updated on September 18, 2026. Enabling Claude in Copilot is therefore a compliance decision to be discussed with the Data Protection Officer (DPO) and the IT department, not a simple setting. The workbook selection tool weighs these criteria according to your context; many large organizations end up using both.
3. Delegating a task: the five-block briefing method
As all publishers say, the richness of the context matters more than the wording. An agent given a vague task will produce a vague deliverable, and the time saved will vanish into revisions. The kit offers a five-block grid developed in our workshops; it's a pedagogical structure, not a publisher standard. The binder generator creates the complete brief based on your answers.
| Bloc | The question to ask oneself | Example |
|---|---|---|
| 1. Background | Who am I, for whom is this work, and why now? | I am a management controller, I am preparing the monthly review for the Executive Committee meeting on Thursday. |
| 2. Deliverable | What should I have in my hands at the end, and in what form? | A 2-page Word document and an Excel spreadsheet showing discrepancies greater than 5%. |
| 3. sources | Which files, folders, exchanges, time periods? | The "Budget 2026" file and the export of the completed work as of the end of September. |
| 4. Constraints | What not to do, the tone, the format, the house rules | Do not send anything, do not modify the source files, amounts in k€. |
| 5. Finishing and checks | How do I know it's over? When should the officer stop to ask me? | It's over when every discrepancy has an explanation; ask me if any data is missing. |
Three reflexes change everything: ask the agent to rephrase the task and ask their questions before starting, which allows you to correct a plan in 30 seconds instead of a deliverable in 30 minutes; give them an example of a successful deliverable; work in a dedicated folder that contains only what the task needs, with a backup copy of important files.
4. 24 tasks to delegate, by function
The kit offers 24 missions drawn from examples published by Anthropic and Microsoft, as well as from our workshops in large organizations. Each mission includes its deliverable, risk level, expected human validation, the most suitable tool, and, in the binder, a ready-to-copy brief. A few examples:
| Function | Mission | Deliverable | Risque |
|---|---|---|---|
| Management, assistance | Daily briefing: emails, Teams, and yesterday's agenda | Summary planned each morning | Low |
| Finance | Analysis of budget and actual variances | Word document and table of discrepancies explained | Medium |
| HR | Preparing for the arrival of a new employee | Integration plan, welcome email, open points | Low |
| Legal | Compare two versions of a contract or policy | Table of annotated differences | Medium |
| Purchasing | Compare the supplier offers received | Comparative grid and clarification questions | Medium |
| Sales | Preparing for a client meeting | Account details and topics to be covered | Low |
| Marketing, communication | Monthly competitive intelligence | Sourced comparative rating | Low |
| Project Manager | Consolidate the team's scattered updates | Weekly progress report | Low |
| IT support | Constructing a timeline of an incident | incident report dated | Medium |
Where to start: a low-risk task that you do at least once a week and that takes you more than 30 minutes. Do it three times in supervised mode, measure it, and only then plan it.
5. Safeguards: the traffic light
An agent acting on your behalf inherits your rights, and their mistakes become yours: Anthropic reminds you that the user remains responsible for all actions Claude takes for them. Payments and contractual commitments, deletions and actions difficult to reverse, automated external communications, decisions concerning individuals, and files containing login credentials, health information, or banking data should never be delegated without genuine human oversight. The traffic light system in the binder asks eight questions and indicates the color of your task.
| Color | Situation | To behave |
|---|---|---|
| Vert | The agent reads non-sensitive internal sources and produces a draft that you review. | Delegate, review, measure |
| Orange | The agent is preparing an action (email, meeting, modification of a shared file) or accessing confidential internal data. | Keep track of approvals at each step, work in a dedicated folder |
| Red | Sensitive personal data, decision concerning an individual, irreversible action, payment | Do not delegate, or only provide anonymized preparation. |
The inherent risk for agents is prompt injection: malicious text hidden in a web page, email, or received document can attempt to issue commands to the agent. Anthropic classifies this risk as non-zero. The countermeasures are simple: limit the sources to what the task requires, maintain validation before any submission, install only connectors and plugins validated by the organization, and do not have the agent navigate to unknown websites while handling sensitive data.
6. For the client: what CERT-FR, the CNIL and the AI Act say
Opening a tool that operates on workstations and within email is not a typical equipment decision. In its bulletin CERTFR-2026-ACT-016 of April 13, 2026, on agentic AI automation products for workstations, the ANSSI CERT-FR cites Claude Cowork and OpenClaw, and recommends limiting personal autonomous assistants like OpenClaw to isolated test environments, without sensitive data. The CNIL and the AI and Digital Council's note of July 20, 2026, emphasizes minimizing access, tracking tasks, and ensuring genuine human validation. Article 4 of the AI Act requires deployers to implement AI control measures among their employees.
These guidelines don't explicitly state "forbidden." They state: no uncontrolled deployment, minimal access rights, a structured pilot phase, and validation by the IT department and the CISO. The toolkit lists the ten decisions to be made before deployment, and who makes the final call; the deployment checklist in the binder details 18 tracked actions with status and progress.
| Decision | Question | Who decides? |
|---|---|---|
| Tool and license | Claude, Copilot Cowork Or both? For whom? | Sponsor, IT department, purchasing |
| Permitted data | What classifications are permitted in the pilot? | DPO, CISO |
| Connectors and plugins | Which ones are open, for reading or writing? | CIO, CISO |
| Prohibited actions | External sending, deletions, browsing: what is being blocked? | RSSI |
| Models | Is Claude activated in Copilot, outside the EU Data Boundary? | DPO, DSI |
| Traceability | Where can we read the activity: Purview, Compliance API, OpenTelemetry? | IT department, compliance |
| Social dialogue | Should the Works Council be informed or consulted? | HR managers, lawyers |
7. Measure the time saved and launch a 30-day pilot program
Without metrics, a pilot's performance is judged by the enthusiasm of the first few days. The toolkit proposes a four-column method: task time without AI, briefing time, review and correction time, and deliverable quality on a scale of 1 to 4. The actual gain is the time without AI minus the briefing and review time; the binder log performs the calculation and highlights the executions that wasted time. The right indicator for the client is not the number of users, but the number of recurring tasks reliably delegated and the net time saved per week on these tasks.
| Period | The goal | Key actions |
|---|---|---|
| Before day 1 | Frame | Choose the tool and the population, define the authorized data and the prohibited actions, configure the administration, inform the CSE if necessary. |
| Week 1 | Lancer | 2-hour onboarding workshop: briefing in five blocks, safeguards, first low-risk mission for each person. |
| Week 2 | Practice | Each participant delegates two recurring tasks in supervised mode and fills in the log of time saved. |
| Week 3 | Share | Peer review: the best briefs become team models, failures are analyzed. |
| Week 4 | Decide | A quantified assessment, reliable missions to plan, rules to adjust, a decision to extend or stop. |
8. What the kit contains
The Word document has 17 pages in 11 chapters: from chat to coworking; Claude Cowork in October 2026; Copilot Cowork in October 2026; Claude Cowork or Copilot Cowork, how to choose; the briefing method; 24 missions to delegate, by function; the safeguards; opening the cowork with confidence for the sponsor; measuring the time saved; the 30-day start-up plan; the resources.
The Excel workbook "Tools from the kit" Cowork » resumes the approach with seven tools: the library of 24 missions with a ready-to-copy brief for each; the brief generator in five blocks; the traffic light “can I delegate?” in 8 questions; the choice between Claude Cowork and Copilot Cowork by weighted criteria, with a score out of 100; the sponsor's deployment checklist in 18 actions, with progress rate; the delegation and time saved log, with summary; the 30-day pilot plan, the deadlines of which are calculated from the launch date.
The kit, chapter by chapter: what each chapter contains and what its purpose is. The templates to fill in are in the Word document and the Excel workbook.
Chapter 1. From chat to coworking: what's changing
Chat, cowork and team agent compared, the simple rule to distinguish them, and what the first uses published by Anthropic, Microsoft, Epoch AI and Ipsos show.
Chapter 2. Claude Cowork in october 2026
What it is, the key dates from January to September 2026, the announced merger with the chat, how it works (isolated space, instructions, connectors, skills, plugins, scheduled tasks) and what the administrator controls in Team and Enterprise.
Chapter 3. Copilot Cowork in october 2026
Integrated validations, skills, automated tasks, the Microsoft 365 scope, licensing and Copilot Credits, administrative controls and the case of Anthropic models in Europe.
Chapter 4. Claude Cowork or Copilot Cowork : how to choose
The comparative table, criterion by criterion, and the five questions to ask yourself before choosing.
Chapter 5. Delegating a task: the briefing method
The five-block grid, three reflexes that change everything and a complete brief ready to adapt.
Chapter 6. 24 tasks to delegate, by function
Management, finance, HR, legal, purchasing, sales, marketing, project management and IT: 24 missions with their deliverables and risk levels, and where to start.
Chapter 7. Safeguards
What we never delegate without real human control: the traffic light, the prompt injection, and the right reflexes on a daily basis.
Chapter 8. For the client: opening the coworking space with confidence
The position of CERT-FR, CNIL and the AI and Digital Council, what the AI Act requires, and the ten decisions to be made before opening.
Chapter 9. Measuring the time saved
The four-column method, the published benchmarks, and the right indicator for the client.
Chapter 10. 30-Day Startup Plan
The week-by-week plan for a pilot of 10 to 30 people, and the five mistakes we see most often.
Chapter 11. Resources
For further information, see the official sources of Anthropic and Microsoft, the positions of the authorities and the studies cited.
Download the kit Cowork (Word + Excel, free)
17 pages in 11 chapters, plus a binder of 7 tools. Written by NextStart.AI based on official Anthropic and Microsoft documentation, the positions of CERT-FR and the CNIL, and workshops we conduct in large organizations.
- Claude Cowork and Copilot Cowork compared, and the tool for choosing
- The briefing method in five blocks and 24 missions with their ready-to-copy briefs
- The traffic light "can I delegate?" and the safeguards
- The sponsor's checklist, the time saved log, and the 30-day pilot plan
9. Frequently Asked Questions
What's the difference between a chat room and a coworking space?
With a chat, you ask a question and copy the answer elsewhere. With a coworking tool, you delegate a complete task: the agent creates a plan, works with your files and applications in several stages, gets back to you at key moments, and delivers a finished product—a document, spreadsheet, or presentation. A recurring team process, with assigned permissions and ownership, is better suited to an agent built in Copilot Studio or Agent Builder.
Claude Cowork or Copilot Cowork Which one to choose?
It depends first on where your data resides. Copilot Cowork It works exclusively within Microsoft 365 (Outlook, Teams, OneDrive, SharePoint, Office) and is suitable for "all-Microsoft" organizations. Claude Cowork It also works with local files and numerous applications, and is suitable for rich analyses and deliverables. The licensing model is also a factor: a per-seat subscription on one hand, and a Copilot license plus usage credits on the other. Many large organizations use both.
Can Claude be used in Copilot? Cowork in Europe ?
Yes, but not by default. In the European Union, EFTA, and the United Kingdom, Anthropic templates are disabled by default in Microsoft 365 Copilot and excluded from the EU Data Boundary, according to Microsoft documentation updated on September 18, 2026. Without administrator action, Copilot Cowork It works with the other available models. Activating Claude is a compliance decision to be discussed with the DPO and the IT department.
What do ANSSI and CNIL say about these tools?
The CERT-FR bulletin CERTFR-2026-ACT-016, published on April 13, 2026, cites Claude Cowork and OpenClaw, and recommends limiting personal autonomous assistants like OpenClaw on workstations to isolated test environments, without sensitive data. The CNIL and AI and Digital Council's note of July 20, 2026, emphasizes minimizing access, traceability, and genuine human validation. These positions do not explicitly state "prohibited": they call for minimal rights, a structured pilot program, and validation by the IT department and the CISO.
What should you never delegate to an agent?
Nothing irreversible or sensitive can happen without real human oversight: payments and contractual commitments, file or data deletions, automated external transmissions, decisions affecting individuals such as recruitment or performance reviews, identification files, health or banking data. The toolkit's traffic light system asks eight questions and classifies the task as green, orange, or red, along with the appropriate course of action.
Is the kit up to date?
The information was verified on October 12, 2026, using official Anthropic and Microsoft sources. It takes into account the merger of Cowork and the cat in "Claude" announced by Anthropic on September 16, 2026. These products evolve every week: check the licensing and configuration points before making any decision.
Who is the kit for?
For employees of large organizations who want to delegate entire tasks to AI, for managers who want to make it a team practice, and for sponsors (business units, CIOs, CISOs) who need to decide how to open Claude Cowork or Copilot CoworkNo technical skills are required.
Is the kit free?
Yes. The Word document and Excel workbook are free, in exchange for a professional email address which will only be used to send you the download link. No unsolicited emails.
10. Going further
This kit complements the resources Cowork from the site: the examples of missions Claude Cowork by job detail missions ready to be copied, the Claude comparison CoworkCopilot Cowork and ChatGPT Work delves deeper into the choice of tool and the diagnostic Cowork situates your organization. Regarding security and rules of use, the Security and data kit and the AI charter template set the framework. Teams that want to learn to delegate in practice do so in the Claude's training Cowork or during a Cowork Party.
Read also : Claude Cowork, Copilot Cowork, ChatGPT Work: the comparison, Configure Claude Cowork the complete guide, Claude Cowork Examples of tasks by profession et AI Project Sponsor Security and Data Kit.
The entire series of free kits. To drive adoption: AI charter template, AI training plan template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Copilot Studio kit, Claude Skills kit et Vibe coding kitFor the AI project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, AI and Social Dialogue Kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit et communicator's AI kit.
11. sources
The information cited on this page and in the kit is based in particular on the following sources, verified on October 12, 2026. The kit provides the complete list in its Resources chapter.
- Anthropic, Cowork becomes ClaudeSeptember 16, 2026; documentation Cowork ; use Claude Cowork safe.
- Anthropic, how people use Claude Cowork ; Compliance API extended to Cowork.
- Microsoft, Copilot general availability CoworkJune 16, 2026; Copilot FAQ Cowork ; Copilot governance and administration Cowork.
- Microsoft, Anthropic templates in Microsoft 365 Copilot ; Microsoft Purview and Copilot Cowork.
- CNIL and the AI and Digital Council, note on agentic AI, July 20, 2026.
- CERT-FR, bulletin CERTFR-2026-ACT-016 of April 13, 2026, reprinted by the Moderator's Blog.
- Epoch AI and Ipsos, delegation of work to AI, August 2026.