Claude Cowork for Development Professionals: A Practical Task Automation Tutorial

Learn how to use Claude Cowork to automate repetitive development work, from programme briefings and research monitoring to grant tracking, partner reporting and knowledge management.

This is a hands-on tutorial. Instead of simply explaining what Cowork can do, you will build a practical workflow, test it, improve it and then turn it into a recurring task.

The examples are designed for programme managers, development professionals, humanitarian practitioners, researchers, policy professionals, grants and partnerships teams, communications staff and M&E professionals.

What You Will Build

By the end of this tutorial, you will have built a simple Cowork workflow that:

• works from a defined set of project information;
• follows clear instructions;
• produces a structured professional briefing;
• flags information requiring human verification;
• does not modify or publish anything without your instruction;
• can later be adapted into a recurring workflow.

No coding is required.

Before You Start

For your first exercise, prepare a small set of non-sensitive project documents. You can use fictional files if you are learning Cowork for the first time.

• A recent project update
• A workplan
• A previous programme briefing
• A project risk or issue log
• Any other document normally used to prepare a weekly update

Anthropic’s current documentation describes Cowork across supported desktop, web and mobile experiences, while some capabilities remain dependent on the desktop application. Anthropic: Get started with Claude Cowork

Step 1 — Choose a Task Worth Automating

Do not start by asking, “What can Cowork do?”

Start with work you already do.

Think about your last reporting cycle, project meeting or research update. Which part took time but did not require much professional judgment?

Good first tasks usually involve collecting, organizing, comparing, summarizing or formatting information.

For this tutorial, we will build a weekly programme briefing.

Step 2 — Open Claude Cowork

Open Claude and select Cowork. The exact interface can vary by platform and plan, so use Anthropic’s current documentation if the options in your account look different.

For the first exercise, keep the task simple. Do not connect every application you use. Give Cowork only the access and information it needs.

Open Anthropic’s current Cowork guide

Step 3 — Define the Outcome

Your first instruction should describe the result you want. Avoid vague requests such as “help me with my project.”

Start with this:

I want you to help me prepare a weekly programme briefing.
Use only the project information I provide for this task.
Review the documents and identify:
- activities completed
- progress since the previous update
- upcoming activities
- important risks or issues
- decisions that may require management attention
Create a concise two-page briefing.
For important factual statements, retain the source.
Do not invent information.
Do not modify the original files.
Do not send or publish anything.
Before finishing, list any information that requires
human verification.

Do not schedule this task yet. First make sure the workflow works when you run it yourself.

Step 4 — Add the Source Material

Now provide the documents Cowork needs.

For this exercise, use:

• the latest project update;
• the previous project update;
• the workplan;
• the risk or issue log.

The important principle is source control. If you want the briefing based only on approved project information, say so explicitly.

Use only the documents provided for this task.
Do not fill gaps with assumptions.
If information is missing, say that it is missing.
If two documents contain different figures or dates,
flag the difference rather than choosing one silently.

This is a simple instruction, but it establishes an important boundary around the evidence Cowork should use.

Step 5 — Run the Workflow

Let Cowork process the material. Then check whether it followed the workflow before worrying about whether the writing sounds polished.

Check five things

1. Did it use the correct documents?
2. Did it identify the correct reporting period?
3. Did it distinguish facts from interpretation?
4. Did it identify missing or contradictory information?
5. Did it avoid changing or sending anything?

If something is wrong, improve the workflow rather than simply correcting the final document yourself.

Step 6 — Improve the Output

Your first workflow may not produce the format you want. That is part of the process.

If Cowork produces a long narrative when you need a management briefing, give it a clearer structure:

Revise the output.
Use this structure:
1. Executive summary
2. Progress this week
3. Upcoming activities
4. Key risks and issues
5. Decisions required
6. Information requiring verification
Keep the final briefing under two pages.
Use short paragraphs and bullet points where appropriate.
Do not add information that is not supported by
the source documents.

The skill is not finding a magic prompt. It is learning to design a workflow that produces a useful and checkable result.

Step 7 — Add Human Review

Before scheduling the workflow, decide what you need to verify yourself.

• Dates are correct.
• Figures match the source documents.
• Programme status is accurately represented.
• Risks have not been overstated or understated.
• Names and organizations are correct.
• Missing information is clearly identified.
• No confidential information has been unintentionally included.

Step 8 — Turn It Into a Recurring Task

Once the workflow works reliably, you can schedule it.

Example: Every Monday at 08:00, prepare the weekly programme briefing.

Anthropic’s current scheduled-task feature allows recurring tasks to run automatically or on demand. Scheduled tasks are available on paid Cowork plans and run remotely, so your computer does not need to stay awake for the scheduled task itself. :contentReference[oaicite:18]{index=18}

Example scheduled instruction

Prepare the weekly programme briefing every Monday.
Use the approved project information available to you.
Review information from the previous seven days.
Produce:
1. Executive summary
2. Progress
3. Upcoming activities
4. Risks and issues
5. Decisions required
6. Information requiring verification
Do not invent information.
Do not modify source documents.
Do not send or publish the briefing.
Clearly identify information requiring human review.

See Anthropic’s current scheduling guide

Step 9 — Build a Research Monitoring Workflow

Now adapt the same workflow-design approach to a weekly research and policy digest.

Prepare a weekly research and policy briefing on [TOPIC].
Search for substantive new developments from the
previous seven days.
Prioritize:
- peer-reviewed research
- government sources
- UN and multilateral organizations
- established research institutions
- credible development organizations
For each relevant item provide:
1. Title
2. Organization or authors
3. Date
4. Source
5. Main finding
6. Why it matters for our work
Exclude promotional material and low-quality sources.
Do not present speculation as fact.
Clearly distinguish source findings from your interpretation.
Create a concise briefing for a development professional.

This can become a recurring task once you have tested the quality of the sources and the usefulness of the output. Anthropic lists recurring research as a common scheduled-task use case. :contentReference[oaicite:20]{index=20}

Step 10 — Build a Grant Deadline Monitor

Grant and funding teams often repeat the same deadline-checking process.

Review the funding and grant documents available for
this project.
Identify:
- proposal deadlines
- reporting deadlines
- financial reporting dates
- renewal dates
- required deliverables
- compliance requirements
- eligibility conditions
Create a deadline table with:
Deadline
Requirement
Responsible person
Source
Status
Flag requirements due within the next 30 days.
Do not change the original documents.
If a deadline is unclear, flag it for human review.

The workflow is simple enough to test manually before deciding whether it should run every week.

Step 11 — Build a Partner Report Consolidation Workflow

If several partners submit similar reports, Cowork can help extract common fields and prepare a consolidated draft.

Review the partner reports in the reporting folder.
For each partner extract:
- reporting period
- activities completed
- key achievements
- challenges
- upcoming activities
- decisions or support requested
- important quantitative figures
Create a consolidated partner update.
Keep each partner clearly identified.
Do not combine figures unless the reports use
compatible definitions and reporting periods.
Flag missing, contradictory or inconsistent information.
Do not invent missing figures.

Similar-looking figures are not necessarily comparable. Definitions, units and reporting periods need to be checked before aggregation.

Step 12 — Build a Knowledge Management Workflow

Another practical application is reviewing and organizing project files.

Review the files in this project folder.
Do not delete or modify anything.
Identify:
- duplicate files
- obsolete versions
- unclear filenames
- inconsistent naming
- missing dates
- files that appear to belong together
Create a proposed folder structure.
Create a proposed naming convention.
Create a list of recommended file changes.
Wait for my approval before modifying or moving files.

Start with recommendations rather than automatic file changes. This gives you an opportunity to review the proposed structure before anything is changed.

Step 13 — Add Connectors Carefully

Connectors can allow Claude to access information and take actions in supported external services. Examples include Google Drive, Slack and other connected applications. Claude inherits the permissions of the connected service. :contentReference[oaicite:21]{index=21}

For your first workflow, however, you may not need a connector at all.

If a workflow works with a small set of approved documents, there is no reason to connect your entire organizational workspace just to produce one briefing.

Anthropic: Use connectors to extend Claude’s capabilities

Step 14 — Use Projects for Repeated Work

If the same programme, grant, research project or portfolio generates many Cowork tasks, Projects can provide a dedicated workspace for related files, context, instructions and memory.

Anthropic currently documents Cowork Projects as a desktop feature using local project data. They are not the same thing as cloud-synced Chat projects. :contentReference[oaicite:22]{index=22}

PROJECT: Education Programme
Context:
- programme documents
- current workplan
- partner reports
- research
- approved terminology
Working rules:
- do not invent programme figures
- retain source references
- flag contradictions
- produce drafts for review
Recurring workflows:
- weekly programme briefing
- monthly partner summary
- quarterly portfolio review

Anthropic: Organize tasks with Projects in Cowork

Step 15 — Use Computer and Browser Access With Care

When a suitable connector is not available, Cowork can use browser or computer interaction to work with applications. Anthropic currently describes computer use as a research preview for Pro and Max in the Claude Desktop application on macOS and Windows. :contentReference[oaicite:23]{index=23}

This can be useful for applications that do not have a direct connector, but screen-based interaction is slower and can be less reliable than a direct connection.

Anthropic also notes that browser interaction can expose whatever sensitive information is visible on the screen. Use a separate browser profile and avoid sensitive sites when appropriate. :contentReference[oaicite:24]{index=24}

Anthropic: Let Claude use your computer in Cowork · Anthropic: Use Claude in Chrome safely

Step 16 — Choose the Right Level of Oversight

Cowork can execute multi-step work without asking for approval at every stage, but the appropriate level of oversight depends on the task.

Use stronger human oversight when:

• the task touches sensitive files or accounts;
• the task can modify files;
• the task can send messages;
• the task interacts with external websites;
• the task affects financial or operational decisions;
• an error could have serious consequences.

Anthropic recommends matching oversight to the stakes of the task and specifically highlights prompt injection as a risk for agentic workflows. :contentReference[oaicite:25]{index=25}

Anthropic: Use Claude Cowork safely

Step 17 — Use This Workflow Template

Once you have built several workflows, you will notice that good instructions usually contain the same elements.

TASK
What needs to be done?
SOURCE
What information should Claude use?
PROCESS
What should it check or do?
OUTPUT
What should it produce?
BOUNDARIES
What must it not do?
VERIFICATION
What should be flagged for human review?
CADENCE
When should it run?

Keep this template. It can become the starting point for designing your own Cowork workflows.

Step 18 — Troubleshoot Your Workflow

Step 19 — Practice Exercises

Exercise 1 — Weekly Programme Briefing

Adapt the workflow from this tutorial to one real or fictional programme. Produce a two-page management briefing from approved project information.

Exercise 2 — Research Digest

Choose a development topic you regularly monitor. Build a weekly digest containing substantive developments, source information and a short explanation of relevance.

Exercise 3 — Grant Deadline Monitor

Use fictional grant documents or public funding information. Ask Cowork to identify upcoming deadlines and requirements and flag anything unclear.

Step 20 — Decide What Not to Automate

A good automation strategy is partly about knowing where to stop.

This approach also fits broader responsible-AI guidance. The OECD AI Principles emphasize human rights, privacy, transparency, human agency and oversight, robustness and accountability. :contentReference[oaicite:26]{index=26} UNESCO’s Recommendation on the Ethics of AI similarly emphasizes human oversight, transparency, fairness, privacy and human rights. :contentReference[oaicite:27]{index=27}

Final Cowork Checklist

• Is the task genuinely repetitive?
• Is the desired output clear?
• Have I identified the correct sources?
• Have I told Cowork what not to do?
• Can I verify the result?
• Have I tested it manually?
• Have I checked how it handles missing information?
• Have I checked how it handles conflicting information?
• Does it have more access than it needs?
• Could an incorrect result cause significant harm?
• Should I use stronger approval controls?
• Is scheduling actually useful?

If You Work in M&E, Go Further

This tutorial is intentionally broader than Monitoring & Evaluation. It is designed for development professionals generally.

If your work specifically involves M&E, MEL or MEAL, EvalCommunity has a dedicated Cowork toolkit focused on M&E workflows. Use the broader tutorial above to learn the workflow-design approach, then move to the specialist resource for M&E applications.

Open the Claude Cowork for M&E Toolkit

Build Your AI Skills for Development Work

Sources and Further Reading

• Anthropic — Get started with Claude Cowork
• Anthropic — Schedule recurring tasks in Claude Cowork
• Anthropic — Organize tasks with Projects in Cowork
• Anthropic — Use connectors to extend Claude’s capabilities
• Anthropic — Use Claude Cowork safely
• Anthropic — Let Claude use your computer in Cowork
• Anthropic — Use Claude in Chrome safely
• OECD — AI Principles
• UNESCO — Recommendation on the Ethics of Artificial Intelligence
• EvalCommunity — Claude Cowork for Monitoring & Evaluation
• EvalCommunity Academy — AI for International Development and Humanitarian Practitioners
• EvalCommunity Academy — AI in Monitoring & Evaluation Certificate
• EvalCommunity Academy — AI for M&E Professional Bundle