Master Claude Cowork in One Afternoon

Prefer to read the lessons here?

The download lets Claude teach you interactively: it asks questions, checks your work, and adapts to your chosen scenario. You can also learn the same core skills below, with explanations, exercises, and prompts you can copy into Claude.

Start with fundamentals, then move to connected workflows. The examples follow State Fair Cafe, a fictional restaurant from the course. You can use the supplied practice files or try the self-contained examples here without downloading anything.

Fundamentals

Your goal in this module is to turn a messy handover into a recommendation someone can act on. The sequence matters: understand the situation, inspect the evidence, make a draft, then check it.

1.1 Start with a real task

In the restaurant scenario, you're taking over a stalled rebrand. The previous manager left scattered notes, customer feedback, and unfinished experiments. Your first job isn't to invent a new brand. It's to find out what happened and what needs a decision.

For your own work, pick an equally concrete finish line: a customer reply, a weekly update, or a comparison of three suppliers. “Help with my business” leaves too many decisions open. Name the audience, the input, and what you want to leave with.

Help me prepare a one-page update for the owner of a neighborhood cafe.
We have inherited an unfinished rebrand and need to decide what to do next.
Ask me what information I have before proposing a recommendation.
The final update should separate confirmed facts, open questions,
and decisions the owner needs to make.

Check your result: Claude should ask for missing evidence. If it immediately invents a complete strategy, point out that the task is to investigate first. In the interactive download, saying “start lesson 1.1” begins this exercise with your chosen scenario.

1.2 Give Claude useful context

A request works better when Claude knows what success looks like. Give it a task, relevant evidence, constraints, and an output format. The model still needs your judgment: a confident sentence is not proof that a number or recommendation is correct.

Here's a small practice brief you can use as-is:

Context: State Fair Cafe is a fictional neighborhood restaurant.
Its regulars value a warm, familiar atmosphere. The owner wants to
modernize without losing that feeling. The new branding is not approved.
Task: Draft three questions to ask the owner before recommending changes.
Constraints: Do not invent customer research or assume a budget.
Format: For each question, explain which decision the answer would inform.

A useful question might be “Which parts of the current experience must stay?” That establishes a constraint. “Should we launch a national campaign?” jumps ahead of the brief. If an answer feels generic, add the information that would change the decision rather than asking Claude to “be better.”

1.3 Explore a folder before changing it

With the downloaded course, open the course folder as your working location in Claude Desktop. Start by inspecting the files, rather than asking Claude to reorganize everything. A map of what exists helps you spot missing information and avoid working from an old draft.

Inspect scenarios/restaurant-rebrand/company-context and
scenarios/restaurant-rebrand/inherited-chaos in this course folder.
Do not edit or move any files.
Give me a short map of the available evidence and the files I should
read first. Explain what each file can help us answer.
Flag conflicting dates, missing context, and anything you cannot open.

Expected result: a file map with reasons to read each source. Customer surveys can tell you what respondents said; a manager's notes tell you what that manager believed. Neither automatically proves why sales changed. Open one cited file yourself and confirm the summary matches it.

Claude's interface and access options can vary by account. Follow Anthropic's working-folder setup guide (opens in a new tab) if your screen differs from the course instructions.

1.4 Turn messy notes into an action list

Extraction is different from brainstorming. When extracting, every action should come from the input. Use the short practice note below before trying a larger file:

Turn this practice handover into an action table.
Columns: action, owner, due date, evidence, missing information.
Do not invent owners or deadlines. Write "not specified" when absent.

Note: Maggie wants the revised menu before Friday's meeting.
Jordan will gather the customer comments. The website needs online
ordering, but nobody has chosen a provider or approved a budget.

The third row describes an unresolved requirement, not permission to buy software. Check that distinction before using the table. For the full exercise, replace the practice note with the course's manager handover and customer feedback, then ask Claude to cite the filenames behind each action.

1.5 Manage a task in small passes

A large request is easier to review when it has checkpoints. Ask for an outline before a full document, and review a sample before processing every file. Continue the same conversation while you are refining the same deliverable; start a fresh one when the task or source material changes substantially.

Before writing the cafe update, show me an outline and a checklist.
Separate reading, analysis, drafting, and verification.
Tell me which inputs are missing. Wait for my feedback on the outline
before creating the draft.

Try a follow-up: “Keep the three recommended actions, but shorten the background to two sentences.” This is more useful than “make it concise” because it tells Claude what to preserve. At each checkpoint, compare the work against your original audience and finish line.

1.6 Split independent research into parallel tasks

The course includes several groups of evidence: customer feedback, competitor research, and menu experiments. Those can be reviewed independently, then combined. A final recommendation depends on all three reviews, so do that after the evidence is collected.

Review the restaurant scenario's customer feedback, competitor research,
and menu experiments as three separate evidence groups.
If parallel agents are available, use one per group; otherwise work
through them in sequence.
For each group, report findings, source filenames, and uncertainties.
Then compare the findings. Highlight disagreements instead of smoothing
them over, and propose two decisions for the owner to consider.
Do not change any source files.

Check your result: each finding should have a source you can inspect. “Everyone wants a modern menu” is too broad if the underlying responses are mixed. Parallel work can speed up reading; it doesn't remove the need to reconcile contradictory evidence.

1.7 Create documents people can use

Choose the format for the job. A one-page document helps an owner make a decision. A spreadsheet helps them inspect assumptions. A presentation helps you explain the recommendation in a meeting. Start with one useful artifact before requesting all three.

Here's a spreadsheet exercise using numbers from the fictional cafe's pie experiment:

Create a small spreadsheet comparing these daily pie-sales scenarios:
Current: price $5.99, 12 slices sold.
Price change only: price $7.99, 12 slices sold.
Price and volume change: price $7.99, 15 slices sold.
Use formulas for daily revenue and change versus current revenue.
Label the last two rows as hypothetical scenarios, not forecasts.
Add an assumptions note: we have not established how customers would
respond to the higher price. Show the table here too so I can check it.

Revenue is price × quantity. It is not profit, and a higher price does not guarantee the same sales volume. Open the generated spreadsheet, inspect its formulas, and change the quantity to see whether the totals recalculate. For a written update, include the recommendation, evidence, assumptions, risks, and the decision you need.

1.8 Research with sources you can check

Use web research to answer a specific question, not to collect impressive-looking statistics. For the cafe, a useful question is what to compare before choosing an online-ordering service: recurring fees, transaction charges, payment processing, and integration requirements.

Research what a small restaurant should check before choosing an
online-ordering service. Use current official provider documentation
for factual claims and link to each source.
Separate recurring fees, per-order charges, payment processing,
and implementation requirements. Mark anything you cannot verify.
Do not recommend a winner without asking about our order volume,
existing website, and budget. Include the date of your research.

Check your result: follow the links, confirm that pricing applies to your country and plan, and look for charges missing from the summary. If a page is inaccessible, provide a relevant excerpt rather than accepting a guessed answer. Browser features depend on availability; see Anthropic's browser guidance (opens in a new tab).

1.9 Review the recommendation before sharing it

Now bring the work together. Your update should explain the situation, support a small number of actions, and make the remaining decision clear. Ask for a critique before treating a polished document as finished.

Review my cafe update from three perspectives: the owner approving
spending, the manager doing the work, and a regular customer.
For each perspective, identify one unclear claim or practical concern.
Check numbers against the supplied evidence. Separate factual errors
from differences of opinion. Suggest edits without rewriting yet.

You're ready to move on when: you can trace the main claims to evidence, explain the spreadsheet's assumptions, and name the next action and its owner. If you can't, return to the relevant source instead of adding more persuasive language.

Connected Workflows

The second module takes the same habits into your own work. Start with a repeated task you understand, connect only the information it needs, and test the result before relying on it.

2.1 Connect your tools to Claude

A connector provides access to a supported service. A downloaded file is a snapshot; a connected source can provide information from that service within its permissions. You still need to specify which records and dates matter.

Choose one tool you already use. Review the access requested during setup, then try a small reading task before requesting changes. Available tools and setup vary; consult Anthropic's Cowork guide (opens in a new tab).

Using the connected tool I specify, find the notes for one project
from the last seven days. Tell me which records you can access.
Summarize decisions, unresolved questions, and next actions.
Link to the source records. Do not send messages or change records.
If the connection is unavailable, tell me what export I can provide.

Check your result: verify a decision against the original record. Missing access should appear as a limitation, not as “nothing happened.” You can practice the same exercise with exported notes if you don't have a connector available.

2.2 Choose a plugin for a task you already do

A plugin packages instructions and capabilities around a type of work. Installing one doesn't give it your company's context or automatically connect every tool it mentions. Start with a task such as preparing a meeting brief, rather than installing a collection you have no immediate use for.

Help me evaluate the plugin I have selected for preparing meeting briefs.
Explain its relevant capabilities and what tools or inputs they require.
Suggest one small exercise using sample data.
Tell me what is unavailable in my setup; don't assume every capability
is installed or connected.

Check your result: you should understand what the plugin adds and which parts you still supply. Compare one output with your normal brief. Keep it only if it improves a task that matters to you.

2.3 Use skills and commands deliberately

Skills supply reusable instructions for a kind of task. Commands, where available, give you an explicit way to start a workflow. Browse the commands actually exposed by your installed plugin rather than copying a command name from a different setup.

Try one command on sample data, then compare it with a plain-language request for the same result. Does it ask better questions? Apply a useful structure? Miss information you normally include?

Explain the inputs required by the meeting-brief command available
in my installed plugin. Help me run it with sample information first.
Afterward, identify which parts came from the plugin's instructions
and which parts depended on the context I supplied.

If no suitable command is available, use the request directly. A shortcut is useful only when the workflow behind it produces a result you can trust.

2.4 Customize the instructions to fit your business

Generic output often comes from generic context. Write down the audience, tone, structure, and approval rules you apply repeatedly. Test those instructions on a sample before adding them to an installed plugin using its supported customization method.

Help me write reusable instructions for a weekly cafe owner update.
Audience: Maggie, the owner. Tone: direct, warm, and practical.
Sections: summary, metrics, progress, risks, next steps, decision needed.
Rules: distinguish actual results from projections; don't invent missing
numbers; identify the source and date of each metric.
Show me the instructions and a sample update using placeholder data.
Do not modify installed plugin files until I have reviewed the draft.

Check your result: replace a placeholder with a missing value and try again. Good instructions should produce an explicit gap, rather than a plausible invented number. Keep a copy of the instructions so you can revise them when your process changes.

2.5 Build your first real workflow

Pick something small enough to test this week: compiling a weekly report, preparing for a client meeting, or turning notes into a draft follow-up. Define the inputs, the result, and the point where you review it. A first workflow does not need to run automatically to be useful.

Help me build one repeatable workflow for my actual job.
Ask me these questions one at a time:
1. What task do I repeat every week?
2. What takes the most time within that task?
3. Which files or tools contain the information?
4. What does the finished result look like?
5. What must I review before anyone else receives it?
Then propose a small version we can test with one example.
Don't promise time savings before we measure the test.

For a weekly update, the workflow might be: gather dated source records → extract changes → draft in your template → verify numbers → review and send yourself. Time a normal run and a trial run, including corrections. An attractive draft that takes longer to fix hasn't solved the problem yet.

2.6 Test it again and keep what works

Run the workflow on a second example. A process that succeeds once may depend on unusually tidy inputs. Try missing information, a conflicting date, or a week with no updates and see how it behaves.

Review our workflow after these two trial runs.
List incorrect outputs, missing inputs, and steps that required correction.
Revise the instructions to address those issues.
Create a short checklist I can use on the next run, including when
I should stop and inspect the original sources.

Finish with something reusable: the instructions, a sample input, a checked output, and a review checklist. That's your first working playbook. The interactive course walks you through this process with feedback; the exercises above give you a place to return when you need a particular skill.